Battery health state estimation method and electronic equipment

Abstract

The invention discloses a battery health state estimation method and electronic equipment. The method comprises the following steps: acquiring state data of a target battery; processing the state data to obtain target data; the target data is input into a target model for prediction processing, the health state estimation value of the target battery is obtained, the target model is an optimized neural network model obtained by training and optimizing a neural network model through a data set, the data set comprises multiple sets of battery test data, and the data set comprises multiple sets of battery test data; each group of test data at least comprises the voltage of the battery at the charging end moment, the voltage drop of the battery, the average current during the discharging period of the battery and the health state value of the battery. According to the method and the device, the problem that the accuracy of the health state evaluation of the battery is relatively low because the health state evaluation of the battery generally depends on the charging and discharging cycle data of the homogeneous battery to carry out working condition table look-up in the prior art is solved.

Description

Technical Field

[0001] The present application relates to the field of lithium batteries, and more specifically, to a method for estimating the health status of a battery and an electronic device. Background Art

[0002] As a key energy supplier for modern technology, lithium batteries play a core role in electric vehicles, portable electronic devices, and renewable energy storage systems. However, batteries will inevitably experience performance degradation during use, which directly affects the reliability, safety, and economy of the battery. Therefore, accurate assessment of battery health status is of great significance for optimizing battery management strategies, improving battery efficiency, and preventing potential failures. Battery health status refers to the ratio of the battery's capacity in its current state to its initial (or nominal) capacity, and is a key indicator for measuring the degree of battery aging.

[0003] Traditional battery health status assessment methods are mostly based on static data, using historical charge and discharge cycle data for analysis and modeling. These methods often assume that under the same charge and discharge cycle conditions, battery state degradation follows a consistent pattern. Therefore, they usually use charge and discharge data obtained under the same battery type and the same operating conditions for analysis, and build a lookup system to predict the battery health status value. However, this method has significant limitations, especially when it needs to deal with complex and changeable vehicle operating conditions, and its prediction accuracy is significantly reduced.

[0004] In the prior art, battery health status assessment generally uses a working condition lookup table of the charge and discharge cycle data of homogeneous batteries, which leads to a low accuracy of battery health status assessment. No effective solution has been proposed so far. Summary of the invention

[0005] The main purpose of the present application is to provide a battery health status estimation method and electronic device to solve the problem that the battery health status assessment in the prior art generally uses the charge and discharge cycle data of homogeneous batteries for working condition lookup, resulting in low accuracy of battery health status assessment.

[0006] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for estimating the health status of a battery is provided. The method comprises: obtaining the status data of a target battery, wherein the status data at least includes the voltage of the target battery at the end of charging, the voltage of the target battery within a preset time after the end of charging, and the current data collected from the target battery during discharge within a preset time interval; processing the status data to obtain target data; inputting the target data into a target model for prediction processing to obtain an estimated health status value of the target battery, wherein the target model is an optimized neural network model obtained by training and optimizing a neural network model using a data set, and the data set includes multiple groups of battery test data, and each group of test data at least includes: the voltage of the battery at the end of charging, the voltage drop of the battery, the average current during the battery discharge, and the health status value of the battery.

[0007] Furthermore, the data set is obtained through the following steps: charging the battery according to a preset charging strategy, and after charging is completed, recording the voltage of the battery at the end of charging and the voltage of the battery within a preset time after the end of charging; subtracting the voltage of the battery within a preset time after the end of charging from the voltage of the battery at the end of charging to obtain the voltage drop of the battery; discharging the battery according to a preset discharge strategy, and during the discharge process, collecting the current data of the battery according to preset time intervals, and calculating the average current during the discharge of the battery; after the discharge is completed, recording the discharge capacity of the battery, calculating the ratio of the discharge capacity to the initial capacity of the battery, and obtaining the health status value of the battery; obtaining each group of test data based on the voltage of the battery at the end of charging, the voltage drop of the battery, the average current during the discharge of the battery, and the health status value of the battery; repeating the above steps until the battery life cycle reaches the termination condition, obtaining multiple groups of test data, and obtaining the data set based on the multiple groups of test data.

[0008] Furthermore, the target model is obtained through the following steps: dividing the data set into a training set and a test set; using the training set to train the neural network model to obtain the neural network model to be tested; using the test set to test the neural network model to be tested to obtain the target model.

[0009] Furthermore, before using the training set to train the neural network model, the method also includes: based on the data set, using the voltage of the battery at the end of charging, the voltage drop of the battery, and the average current during the discharge of the battery as input features of the input layer of the neural network model; designing multiple hidden layers in the neural network model, wherein each hidden layer includes at least a preset activation function and a preset number of neurons; and using the health status value of the battery as the output target of the output layer of the neural network model.

[0010] Furthermore, using the training set to train the neural network model includes: initializing the weights and bias parameters of the multi-layer hidden layers and output layers in the neural network model; inputting the voltage of the battery at the end of charging, the voltage drop of the battery, and the average current during battery discharge in the training set into the neural network model, executing the forward propagation algorithm, and obtaining the estimated health state of the battery; using the automatic feature selection algorithm to optimize the loss function to obtain an optimized loss function; quantifying the difference between the estimated health state of the battery and the actual health state of the battery in the training set according to the optimized loss function; based on the difference, using the back propagation algorithm to calculate the gradient of the optimized loss function relative to the weights and bias parameters of the multi-layer hidden layers and the output layer; using the optimization algorithm, updating the weights and bias parameters of the multi-layer hidden layers and the output layer according to the gradient; repeating the above steps until the neural network model reaches the preset training stop condition.

[0011] Furthermore, the test set is used to test the neural network model to be tested, and obtaining the target model includes: inputting the voltage of the battery in the test set at the end of charging, the voltage drop of the battery, and the average current during the battery discharge into the neural network model to be tested, executing the forward propagation algorithm to obtain the estimated health state of the battery; calculating the preset performance index to quantify the difference between the estimated health state of the battery and the actual health state of the battery in the test set; evaluating the generalization ability of the neural network model based on the performance index; repeating the above steps until the performance of the neural network model meets the preset conditions to obtain the target model.

[0012] Further, the status data is processed to obtain the target data, including: obtaining from the status data the voltage of the target battery at the end of charging, the voltage of the target battery within a preset time after the end of charging, and the current data collected from the target battery within a preset time interval; subtracting the voltage of the target battery within a preset time after the end of charging from the voltage of the target battery at the end of charging to obtain the voltage drop of the target battery; calculating the average current of the target battery during discharge based on the current data collected from the target battery within the preset time interval; and using the voltage of the target battery at the end of charging, the voltage drop of the target battery, and the average current of the target battery during discharge as the target data.

[0013] Furthermore, the loss function is optimized by using an automatic feature selection algorithm, and the optimized loss function includes: using the automatic feature selection algorithm to calculate the contribution of the voltage of the battery at the end of charging, the voltage drop of the battery, and the average current during battery discharge in the training set to the prediction of the estimated value of the battery health state, and obtain multiple contribution values; adjusting the automatic feature selection algorithm according to the multiple contribution values ​​to obtain an adjusted automatic feature selection algorithm; optimizing the loss function with the adjusted automatic feature selection algorithm to obtain an optimized loss function.

[0014] Furthermore, after using the training set to train the neural network model to obtain the neural network model to be tested, the method also includes: using an interpretability tool to perform an interpretability analysis on the neural network model to be tested to obtain an interpretability analysis result; based on the interpretability analysis result, adjusting the neural network model to be tested until the interpretability index of the neural network model to be tested reaches a preset standard.

[0015] According to another aspect of the present application, an electronic device is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a method for performing any battery health status estimation method.

[0016] In an embodiment of the present application, by acquiring status data of a target battery, wherein the status data at least includes the voltage of the target battery at the end of charging, the voltage of the target battery within a preset time after the end of charging, and the current data during the discharge period collected from the target battery within a preset time interval; processing the status data to obtain target data; inputting the target data into a target model for predictive processing to obtain an estimated value of the health state of the target battery, wherein the target model is an optimized neural network model obtained by training and optimizing a neural network model using a data set, the data set including multiple groups of battery test data, each group of test data including at least: the voltage of the battery at the end of charging, the voltage drop of the battery, the average current during the discharge of the battery, and the health state value of the battery, the problem that the battery health state assessment in the prior art generally relies on the charge and discharge cycle data of homogeneous batteries for working condition lookup, resulting in low accuracy of the battery health state assessment, the target data obtained based on the status data of the target battery is input into the optimized neural network model, and the optimized neural network model is used to estimate the battery health state value, thereby achieving the technical effect of improving the accuracy of the battery health state assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 A hardware structure block diagram of a computer terminal for implementing a method for estimating a battery health state is shown;

[0019] Figure 2 is a flow chart of a method for estimating a battery health status according to an embodiment of the present application;

[0020] Figure 3 is a flowchart of a method for constructing a data set provided in an embodiment of the present application;

[0021] Figure 4 is a flowchart of a method for constructing a target model provided in an embodiment of the present application;

[0022] Figure 5 is a flow chart of a method for training a target model according to an embodiment of the present application;

[0023] Figure 6 It is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] It should be noted that the collected data involved in this application (including but not limited to the data set, the status data of the target battery, etc.) are authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good customs, and provide corresponding operation entrances for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions to provide users with corresponding operation entrances for users to choose to agree or refuse the automated decision results; if the user chooses to refuse, the expert decision-making process will be entered.

[0027] Example 1

[0028] According to an embodiment of the present application, an embodiment of a method for estimating a battery health status is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0029] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal (or mobile device) for implementing a method for estimating a battery health state. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0030] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0031] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method for estimating the battery health status in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the above-mentioned method for estimating the battery health status is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0032] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0033] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0034] Under the above operating environment, this application provides Figure 2 A method for estimating the health status of a battery is shown. Figure 2 This is a flowchart of a method for estimating a battery health status according to Example 1 of the present application.

[0035] Step S201, obtaining status data of a target battery, wherein the status data at least includes the voltage of the target battery at the end of charging, the voltage of the target battery within a preset time after the end of charging, and the current data of the target battery during discharge collected within a preset time interval.

[0036] Optionally, the above-mentioned status data can be various key parameters of the target battery under specific operating conditions, such as voltage, current, temperature and other data, and the above-mentioned status data can be used to evaluate the current state and health of the target battery. The voltage of the above-mentioned target battery at a preset time after the charging is completed refers to the voltage value within a short time (such as 1 second, 0.1 second, etc.) after the battery is charged, which reflects the transient change of the battery voltage after charging is stopped, which is helpful to evaluate the internal state of the battery, such as electrolyte distribution and charge balance. When the current data during the discharge period of the target battery is collected within a preset time interval, the time interval can be set according to the specific application requirements, battery characteristics and data processing capabilities to achieve the best monitoring effect and data management efficiency, such as 10 seconds, 20 seconds, 30 seconds, 100 seconds, etc.

[0037] Step S202, processing the state data to obtain target data.

[0038] Optionally, the above-mentioned status data cannot be directly used as input data of the neural network model. It needs to be converted into the input format set by the neural network model to obtain the target data before the neural network model can be used to estimate the health status of the target battery.

[0039] For example, the state data A includes the voltage of the target battery at the end of charging, 10V, the voltage of the target battery within a preset time after the end of charging, 9V, and the voltage of the target battery at each time from t1 to t 100 The current values ​​obtained by collecting current data once per second are I k (k=1,2,...,100), the input nodes of the neural network model are the voltage drop of the target battery and the average current during battery discharge. Therefore, the state data A is processed and the voltage drop of the target battery is calculated to be 1V, and the average current during discharge is assumed to be 15V, which together with the voltage of the target battery at the end of charging, 10V, constitute the above target data.

[0040] Step S203, input the target data into the target model for prediction processing to obtain the estimated health status of the target battery, wherein the target model is an optimized neural network model obtained by training and optimizing the neural network model using a data set, and the data set includes multiple groups of battery test data, and each group of test data includes at least: the voltage of the battery at the end of charging, the voltage drop of the battery, the average current during battery discharge, and the health status value of the battery.

[0041] Optionally, the above-mentioned target model refers to a neural network model that can predict the health status of the battery based on specific input features after sufficient training and optimization. The voltage at the end of the battery charging reflects the electrochemical state of the battery in a fully charged state; the voltage drop of the battery, that is, the voltage difference between the end of charging and a short interval, is an important indicator for evaluating the instantaneous charge distribution of the battery; the average current during the battery discharge describes the average power output capacity of the battery during the discharge process, which is a direct indicator for measuring battery performance. The combination of these three input features can provide rich and important information from the perspectives of electrochemistry, energy storage, and dynamic discharge performance, which helps the neural network model learn the characteristics of battery aging and health status changes from multiple dimensions, thereby improving the accuracy of battery health status prediction.

[0042] For example, assume that the adjusted weights and biases of the target model are as follows: weights from the input layer to the first hidden layer: W11=0.3, W12=0.2, W13=0.4, W21=0.4, W22=0.3, W23=0.2; biases from the input layer to the first hidden layer: b1=-0.1, b2=0; weights from the first hidden layer to the second hidden layer: w11=0.2, w12=-0.1, w13=0.1, W21=0.3, W22=0.2, W23=0.1; W31=0.1, W32=0.2, W33=0.3; biases from the first hidden layer to the second hidden layer: b1=0.1, b2=0.2, b3=0.3; weights and biases from the second hidden layer to the output layer: wout=0.8, bout=0.9. The target data include the voltage of the target battery at the end of charging is 0.9V, the voltage drop of the target battery is 0.1V, and the average current of the target battery during discharge is 0.6A. The above target data is input into the input layer of the target model, and the value of the first neuron of the first hidden layer is calculated according to the weight and bias as Z1=0.3×0.90×0.10×0.6-0.1=-0.01; the value of the second neuron of the first hidden layer is Z2=0.4×0.90×0.10×0.6+0.0=0.21; using the ReLU activation function, the output of the first hidden layer is A1=0, A2=0.21; and the output of the second hidden layer is calculated by analogy as A1=0.019, A2=0.098, A3=0.078; the estimated value of the battery health status of the output layer is 0.8×(0.019+0.098+0.078)+0.9=0.956.

[0043] In the embodiment of the present application, by acquiring the state data of the target battery, wherein the state data at least includes the voltage of the target battery at the end of charging, the voltage of the target battery within a preset time after the end of charging, and the current data during the discharge period collected from the target battery within a preset time interval; processing the state data to obtain the target data; inputting the target data into the target model for prediction processing to obtain the estimated health state value of the target battery, wherein the target model is an optimized neural network model obtained by training and optimizing the neural network model using a data set, and the data set includes multiple groups of battery test data, each group of test data includes at least: the voltage of the battery at the end of charging, the voltage drop of the battery, the average current during the battery discharge, and the health state value of the battery, which solves the problem that the battery health state evaluation in the prior art generally relies on the charge and discharge cycle data of homogeneous batteries for working condition lookup, resulting in low accuracy of the battery health state evaluation. In the present application, the target data obtained according to the state data of the target battery is input into the optimized neural network model, and the optimized neural network model is used to estimate the health state value of the battery, thereby achieving the technical effect of improving the accuracy of the battery health state evaluation.

[0044] In an optional embodiment, if Figure 3 As shown, the above data set can be obtained through the following steps:

[0045] Step S301, performing a charging test on the battery according to a preset charging strategy, and after the charging is completed, recording the voltage of the battery at the time when the charging is completed and the voltage of the battery within a preset time after the charging is completed.

[0046] Optionally, the above-mentioned preset charging strategy can be a charging process and parameters, which are used to evaluate the performance of the battery under charging conditions, and may include parameters such as charging rate, charging voltage, charging time, and charging temperature. For example, the preset charging strategy may be to charge the battery to 100% SOC at a fast charging rate of 2C at a constant temperature of 25°C, and leave the battery for 15 minutes. The voltage of the above-mentioned battery at the end of charging is usually the reading when the battery reaches the preset charging cut-off voltage. These data points, the voltage of the battery at the end of charging and the voltage of the battery within a preset time after the end of charging, provide information on the electrochemical behavior of the battery under different states, which is crucial for subsequent neural network model training.

[0047] For example, the battery is first charged with constant current until the battery voltage reaches 4.2V, and then the constant voltage charging stage is switched to, until the charging current drops to 0.05C, and then the charging is terminated. Assume that the interval time set after the charging is completed is 1 second, that is, the preset time is 1 second. The voltage of the battery at the end of charging and 1 second after the charging is completed is recorded, that is, the voltage of the battery at the end of charging is 4.2V, and the voltage of the battery 1 second after the charging is completed is 4.18V.

[0048] Step S302, the voltage drop of the battery is obtained by subtracting the voltage of the battery within a preset time after the charging is completed from the voltage of the battery at the charging completion time.

[0049] Optionally, the voltage drop of the battery is the voltage difference between the voltage of the battery at the end of charging and the voltage of the battery within a preset time after the end of charging. The calculation of the voltage drop of the battery can not only reflect the transient electrochemical behavior of the battery at the end of the charging process, but also reveal the charge distribution state inside the battery, which is a sensitive indicator for evaluating the health status of the battery.

[0050] For example, according to the case data in step S301, it can be calculated that the voltage drop of the battery is 0.02V.

[0051] Step S303, performing a discharge test on the battery according to a preset discharge strategy, collecting current data of the battery at preset time intervals during the discharge process, and calculating an average current of the battery during the discharge period.

[0052] Optionally, the above-mentioned preset discharge strategy can be a discharge process and parameters, which are used to evaluate the performance of the battery under discharge conditions, and may include parameters such as discharge rate, discharge time, discharge depth, and discharge temperature. For example, the above-mentioned preset discharge strategy can be under a constant temperature of 10°C to stabilize the state of the battery cell. Then discharge the battery to a cut-off voltage of 3.0V at a discharge rate of 1C, and then let the battery stand for 30 minutes. When collecting current data, the above-mentioned preset time interval can be 10 milliseconds, 1 second, 10 seconds, etc. During the discharge process, the current data of the battery is collected at preset time intervals (for example, once every 10 milliseconds or once per second). By calculating the average current during the discharge of the battery, the energy output capacity of the battery during the discharge process can be reflected, which is closely related to the evaluation of the health status of the battery.

[0053] For example, in a discharge test, from t1 to t 100 The current data is collected once per second, and the current values ​​obtained are I k (k=1,2,...,100), add up the 100 collected current values ​​and divide by the total number of data points 100 to get the average current during battery discharge.

[0054] Step S304, after the discharge is completed, the discharge capacity of the battery is recorded, and the ratio of the discharge capacity to the initial capacity of the battery is calculated to obtain the health status value of the battery.

[0055] Optionally, the discharge capacity of a battery refers to the amount of charge that a battery can release under a preset discharge strategy, which can be expressed in ampere-hours (Ah) or milliampere-hours (mAh). As the battery is used and ages, the discharge capacity will gradually decrease. The decrease in the battery discharge capacity is one of the main manifestations of the battery's health degradation. By accurately monitoring and analyzing the battery's discharge capacity, a comprehensive assessment of the battery's health status can be achieved.

[0056] For example, the discharge capacity of the battery after discharge is 4.5 Ah, the initial capacity of the battery is 5 Ah, and the calculated health status value of the battery is 90%.

[0057] In step S305, each set of test data is obtained according to the voltage of the battery at the end of charging, the voltage drop of the battery, the average current during the discharge of the battery, and the health status of the battery.

[0058] Optionally, in the process of constructing the test data set, ensure that each data set contains four features: the battery voltage at the end of charging, the battery voltage drop, the average current during battery discharge, and the battery health status value. By combining these features, a test data set that fully reflects the battery health status is formed. Each set of test data represents a snapshot of the battery state under a specific charge and discharge cycle, which can be used to train and verify the neural network model to ensure that the model can accurately predict the battery health status.

[0059] For example, a charge and discharge cycle test was performed on the battery, and the following data were collected during the test: the battery voltage at the end of charging was 4.2V, the battery voltage 1 second after the end of charging was 4.1V, a current data set consisting of current data collected every 10 seconds during the discharge process, the battery discharge capacity was 2000mAh or 2Ah (assuming full discharge to 0% SOC), and the battery initial capacity was 3Ah. Calculation of the above data revealed that the battery voltage drop was 0.1V, the average current during battery discharge was 12A, and the battery health state was 66.67%. Based on the battery voltage of 4.2V at the end of charging, the battery voltage drop of 0.1V, the average current during battery discharge of 12A (assuming), and the battery health state of 66.67%, a set of test data was obtained.

[0060] Step S306, repeat the above steps until the battery life cycle reaches the termination condition, obtain multiple groups of test data, and obtain a data set based on the multiple groups of test data.

[0061] Optionally, the battery life cycle termination condition may be that the capacity of the battery cell decays to less than 80% of the initial capacity, the power output capacity of the battery cell is less than a certain threshold, etc. By performing multiple charge and discharge test cycles on the battery, multiple groups of test data covering different health states of the battery can be collected, and a data set containing different health states of the battery during the charge and discharge process can be constructed.

[0062] Through the above technical solution, the three key characteristics of the battery, namely voltage drop, average discharge current and end-of-charge voltage, are collected to construct a test data set reflecting the multi-dimensional state of the battery, providing a key data basis for the neural network model to predict the health status of the battery.

[0063] In an optional embodiment, if Figure 4 As shown, the target model can be obtained through the following steps:

[0064] Step S401, dividing the data set into a training set and a test set.

[0065] Optionally, the above data set can be divided according to a certain ratio, for example, 80% of the data is used as a training set for model training, and the remaining 20% ​​of the data is used as a test set to evaluate the generalization ability of the model. This division method ensures that the amount of data for model training is sufficient, while also retaining enough data to objectively evaluate the performance of the model, avoiding the problem of model overfitting.

[0066] Step S402: Use the training set to train the neural network model to obtain a neural network model to be tested.

[0067] Optionally, the goal of training the neural network model is to adjust the weights and biases of the neural network during the training process to minimize the difference between the predicted battery health status value and the actual battery health status value. After the training is completed, the above-mentioned neural network model to be tested is obtained.

[0068] Step S403: Use the test set to test the neural network model to obtain a target model.

[0069] Optionally, after obtaining the above-mentioned neural network model to be tested, the model is tested using the test set divided in step S401. The test set contains new data that the model has never seen, and can objectively evaluate the prediction accuracy of the model. If the performance of the model on the test set meets the preset standards, then the model is considered to be a target model that can accurately predict the health status of the battery.

[0070] Through the above technical solution, the training set and test set are reasonably divided and applied, which not only realizes the effective training of the neural network model, but also verifies the prediction performance of the model, ensuring the reliability and accuracy of the model in the health status estimation of the actual battery.

[0071] In an optional embodiment, before using the training set to train the neural network model, the method further includes:

[0072] In the first step, based on the data set, the battery voltage at the end of charging, the battery voltage drop, and the average current during battery discharge are used as input features of the input layer of the neural network model.

[0073] Optionally, the input layer can be designed according to the specific application scenario and the characteristics of the input data.

[0074] For example, the input layer of the lithium battery health status estimation model contains three input nodes, among which input node 1 is the voltage of the battery at the end of charging, and receives the final voltage value during the charging process as data input; input node 2 is the voltage drop of the battery, and receives the degree of voltage drop after charging is completed as data input; input node 3 is the average current during battery discharge, and receives the average current intensity during the discharge process as data input.

[0075] The second step is to design multiple hidden layers in the neural network model, wherein each hidden layer includes at least a preset activation function and a preset number of neurons.

[0076] Optionally, the hidden layer is located between the input layer and the output layer. In a neural network, multiple hidden layers are usually set. Each hidden layer consists of multiple neurons. Neurons and layers are connected by weights and biases. The design of multiple hidden layers allows the model to learn more complex features and patterns in the data through layer-by-layer nonlinear transformations, thereby showing better performance in many tasks. Among them, the activation function and the number of neurons in the hidden layer are optional. The activation function of each hidden layer is used to transform the output of the previous layer into the input of the next layer, which can provide the model with the ability to handle nonlinear problems and ensure that the model can capture the nonlinear mapping relationship between the input features and the health status of the battery. The activation functions that can be used include ReLU, tanh, sigmoid, etc.; and the number of neurons directly affects the learning ability and generalization performance of the model, so the above-mentioned preset number of neurons can be determined according to the complexity of the problem and the characteristics of the data set.

[0077] For example, suppose a deep neural network is constructed for accurate estimation of the health status of lithium batteries. Considering that the prediction of battery health status is a relatively complex process, a neural network with three hidden layers is designed, in which the first hidden layer selects ReLU as the activation function (because the battery feature data may contain a large number of positive values, ReLU can effectively handle this situation), and the number of neurons in the first hidden layer is set to 64, which can handle relatively complex feature interactions without causing overfitting too quickly; the second hidden layer continues to use ReLU as the activation function, and the number of neurons is set to 128, which enables the model to further learn higher-level feature representations and increase the complexity of the model to adapt to possible complex relationships; the third hidden layer takes into account the increase in model depth, and the ReLU activation function can be selected again, and the number of neurons is set to 256, aiming to maximize the learning potential of the model.

[0078] The third step is to use the battery health status value as the output target of the output layer of the neural network model.

[0079] Optionally, the output layer of the above neural network model is the last layer of the entire neural network model, which is responsible for converting the model's prediction into the final output form. The above output target is the expected output value during model training and prediction. Taking the battery health status value as the output target means that the training process of the neural network model will focus on learning how to predict the battery health status value based on the input features.

[0080] Through the above technical solution, a neural network model that can effectively predict the health status of the battery is constructed. By designing a neural network architecture containing multiple hidden layers for learning, the technical effect of improving the accuracy of battery health status prediction is achieved.

[0081] In an optional embodiment, if Figure 5 As shown, using the training set to train the neural network model includes:

[0082] Step S501, initializing the weights and bias parameters of the multiple hidden layers and output layers in the neural network model.

[0083] Optionally, the above initialization method can be zero initialization, random initialization, Xavier / Glorot initialization, He initialization, orthogonal initialization, identity initialization, etc. The above weights are parameters connecting different neurons in the neural network, which are used to indicate the contribution of input features to the output results. In the process of transmitting information between neurons, the input value on each connection will be multiplied by the corresponding weight. The size of the weight determines the influence of the input feature on the output. A larger weight indicates that the feature has a greater influence on the output, and a smaller weight indicates a smaller influence. The above bias is another parameter in the neural network, which is similar to the intercept term in mathematics and is used to adjust the activation point of the neuron. The bias parameter exists separately in each neuron, and its function is to enable the neuron to be activated when the input feature is 0, or it provides the initial value of the neuron output. The setting of the bias allows the model to fit the data better because it allows the neuron to have a non-zero output even without input, thereby increasing the flexibility of the model. The purpose of initializing the weight and bias parameters is to provide a starting point for the network. From this starting point, the neural network model can gradually adjust the parameters through the training process to minimize the prediction error.

[0084] For example, the neural network model contains 3 hidden layers, each with 128 neurons, and one neuron in the output layer. The weights are initialized using the Xavier initialization method. For a hidden layer with 128 neurons, the weight matrix is ​​a 128x128 matrix, and each element is initialized to a value randomly drawn from a normal distribution (mean 0, standard deviation 1 / sqrt(128)) to ensure good gradient propagation. The bias parameter is initialized to a zero vector, which helps the model respond to input data closer to the natural state in the early stages of training.

[0085] Step S502, inputting the battery voltage at the end of charging, the battery voltage drop, and the average current during battery discharge in the training set into the neural network model, executing the forward propagation algorithm, and obtaining the estimated value of the battery health state.

[0086] Optionally, the above-mentioned forward propagation algorithm is a basic calculation process in a neural network, which is used to calculate the output of each layer in sequence starting from the input layer of the network until the final output layer result is obtained. This process is based on the connection weights and biases between neurons in the network, as well as the activation function. After the input layer receives the input data, the input data is propagated and processed through the hidden layer. The neurons in each layer are calculated based on the output of the neurons in the previous layer and the weights connected to them to obtain the weighted input sum of the current neuron, and the weighted input sum of the neuron is converted into the output of the neuron through the activation function.

[0087] For example, assume that the adjusted weights and biases of the target model are as follows: weights from the input layer to the first hidden layer: W11=0.3, W12=0.2, W13=0.4, W21=0.4, W22=0.3, W23=0.2; biases from the input layer to the first hidden layer: b1=-0.1, b2=0; weights from the first hidden layer to the second hidden layer: W11=0.2, W12=-0.1, W13=0.1, W21=0.3, W22=0.2, W23=0.1; W31=0.1, W32=0.2, W33=0.3; biases from the first hidden layer to the second hidden layer: b1=0.1, b2=0.2, b3=0.3; weights and biases from the second hidden layer to the output layer: wout=0.8, bout=0.9. The battery test data group B includes a battery voltage of 0.9V at the end of charging, a battery voltage drop of 0.1V, and an average current of 0.6A during battery discharge. The above data is input into the input layer of the target model for forward propagation. The value of the first neuron of the first hidden layer is calculated according to the weight and bias as Z1=0.3×0.90×0.10×0.6-0.1=-0.01; the value of the second neuron of the first hidden layer is Z2=0.4×0.90×0.10×0.6+0.0=0.21; using the ReLU activation function, the output of the first hidden layer is A1=0, A2=0.21; and the output of the second hidden layer is calculated to be A1=0.019, A2=0.098, A3=0.078 by analogy; the battery health status estimation value of the output layer is 0.8×(0.019+0.098+0.078)+0.9=0.956.

[0088] Step S503, optimizing the loss function using an automatic feature selection algorithm to obtain an optimized loss function.

[0089] Optionally, the above loss function is used to quantify the gap between the estimated value of the neural network model and the actual value. The loss functions that can be used include mean square error, mean absolute error, cross entropy loss, etc. The above automatic feature selection algorithm can identify the input features that contribute most to the model prediction, reduce the features that have little or no effect on the prediction results, and thus optimize the loss function. This optimization process can improve the sensitivity and accuracy of the neural network model prediction, improve model performance, reduce the risk of overfitting, and reduce computational costs. The automatic feature selection algorithm can be L1 regularization, L2 regularization, etc.

[0090] For example, assuming that the initial loss function is a mean square error (MSE), during the training process, an automatic feature selection algorithm L1 regularization is used to optimize the mean square error loss function. It will add the sum of the absolute values ​​of all weights as a penalty term in the loss function, penalizing hidden layer neurons whose weights are greater than a certain threshold, thereby reducing the complexity of the model without sacrificing the model's predictive ability, helping to identify the features that contribute most to the prediction of the battery's health status.

[0091] Step S504, quantifying the difference between the estimated value of the health state of the battery and the actual value of the health state of the battery in the training set according to the optimized loss function.

[0092] For example, the actual value of the battery health status in the battery test data group A in the training set is 0.83 (i.e. 83%). The input features in the battery test data group A are input into the neural network model for prediction processing, and the estimated value of the battery health status is 0.85. Based on the loss calculation of MSE, the difference, that is, the loss, is (0.850.83)^2=0.0004. Combined with L1 regularization, assuming that there are 10 weight parameters, the average absolute value of all weight parameters is 0.1, that is, the sum S=10×0.1=1. If the regularization parameter λ=0.01, the loss of the L1 regularization part is 0.01×1=0.01. In this way, the total loss for data group A is 0.0004+0.01=0.0104.

[0093] Step S505, based on the difference, use the back propagation algorithm to calculate the gradient of the optimized loss function relative to the weights and bias parameters of the multiple hidden layers and the output layer.

[0094] Optionally, the back-propagation algorithm described above can be used to minimize the loss function, which provides guidance for adjusting these parameters by calculating the gradient of the loss function with respect to each weight and bias parameter so that the predicted output of the model is closer to the actual value, thereby improving the accuracy of the model's prediction.

[0095] For example, suppose the neural network model includes an input layer, two hidden layers (128 neurons in each hidden layer), and an output layer. The selected loss function is the mean square error (MSE). Assuming the weight of the output layer is W3 and the bias is B3, calculate dMSE / dW3 and dMSE / dB3. Similarly, the gradient is propagated backward to the second hidden layer, and the gradient of the loss function relative to W2 and B2 is calculated, and then to the first hidden layer, the gradient of the loss function relative to W1 and B1, dMSE / dW1 and dMSE / dB1, is calculated.

[0096] Step S506: Use an optimization algorithm to update the weights and bias parameters of the multiple hidden layers and the output layer according to the gradient.

[0097] Optionally, the above optimization algorithm can be used to find the optimal configuration of model weights and bias parameters, and then adjust the parameters of the neural network model. It can be stochastic gradient descent SGD, Adam optimization algorithm, SGD with momentum, Adagrad algorithm, etc.

[0098] For example, using stochastic gradient descent SGD to update weight and bias parameters, assuming that the η learning rate of SGD is 0.01, the gradients of weight and bias are obtained from back propagation. For the weight W3 of the output layer, the calculated gradient is dMSE / dW3=0.001, and the gradient of bias B3 is dMSE / dB3=-0.0005. Taking the weight W3 of the output layer as an example, if its initial value is 1.2, the updated value is: W3=1.2-0.01*0.001=1.19999; similarly, the update of bias B3 is as follows: B3=B3-0.01*(-0.0005)=B3+0.00005.

[0099] Step S507, repeat the above steps until the neural network model reaches a preset training stop condition.

[0100] Optionally, the above-mentioned preset training stop condition may be reaching a predetermined number of iterations (such as 1000 iterations), the value of the loss function is less than a certain threshold, or the improvement of the loss function is no longer significant, etc. (i.e., the training stop condition may be a fixed number of iterations, or when the rate of change of the loss function is less than a certain threshold). Repeating steps S502 to S506 can ensure that the model parameters are fully optimized so that the model can achieve the highest accuracy in predicting the health status of the battery.

[0101] For example, the training stop condition is set to end when the loss function drops below 0.001 or reaches 1000 iterations, which means that the model will continue to perform steps S502 to S506 until any of the above stop conditions is met.

[0102] Through the above technical solution, the generalization ability of the neural network model is optimized through the automatic feature selection algorithm, overfitting is avoided, and the reliability of the neural network model is improved.

[0103] In an optional embodiment, the neural network model to be tested is tested using a test set to obtain a target model including:

[0104] In the first step, the voltage of the battery in the test set at the end of charging, the voltage drop of the battery, and the average current during battery discharge are input into the neural network model to be tested, and the forward propagation algorithm is executed to obtain the estimated value of the battery's health status.

[0105] For example, the test set contains a battery test data set A, including the battery voltage of 4.15V at the end of charging, the battery voltage drop of 0.05V, and the average current of 10A during battery discharge. This set of data is input into the neural network model to be tested, the forward propagation algorithm is executed, and the output layer calculates the estimated value of the battery health status based on the learned parameters.

[0106] In the second step, the preset performance indicators are calculated to quantify the difference between the estimated health status of the battery and the actual health status of the batteries in the test set.

[0107] Optionally, the above-mentioned preset performance indicators can be used to quantify the difference between the model prediction result and the actual target value, so as to determine whether the model is effective and to what extent. The performance indicators can be mean square error (MSE), mean absolute error (MAE), determination coefficient (R 2 coefficient), etc.

[0108] For example, assuming that the estimated health status of the battery obtained in the first step is 0.87, and the actual health status of the battery in the battery test data group A in the test set is 0.85, according to the MSE and L1 regularization calculation formula, the loss is calculated to be 0.0104, which is the difference between the estimated value and the actual value.

[0109] The third step is to evaluate the generalization ability of the neural network model based on performance indicators.

[0110] Optionally, the generalization ability mentioned above refers to the prediction performance of the model on unseen data, which is an important indicator to measure whether the model is overfitting. If the performance indicators of the model on the test set are similar to those on the training set, such as MSE, MAE and other indicators are close, then it can be considered that the model has good generalization ability. Otherwise, if the indicators on the test set are significantly higher than those on the training set, this indicates that the model may be overfitting to the training data and has poor prediction ability for new data.

[0111] For example, assuming that the performance preset condition of the neural network model is MSE ≤ 0.0005, then in the second step case, MSE = 0.0004, the generalization ability of the model is preliminarily evaluated as meeting the standard. However, the performance of the entire model is usually evaluated on the test set, not just a single sample, so it is necessary to calculate the average MSE of the entire test set. If the average MSE is lower than the above preset condition, it can be said that the model has good generalization ability.

[0112] In addition, during the evaluation process, interpretability tools can also be combined to provide interpretability proof of model decisions. For example, feature importance graphs generated using SHAP, LIME, etc. can be displayed together with the performance indicators of the model on the validation set to help R&D personnel fully understand the model's predictive capabilities and its decision-making basis. By observing the prediction results and explanations on the validation set, possible limitations or errors in the model can be identified, so that targeted model improvements can be made. For example, if the SHAP value graph shows that voltage drop has a negative contribution to the prediction of SOH, it is common sense that voltage drop should be positively correlated with SOH, which may indicate that the model is biased or wrong.

[0113] The fourth step is to repeat the above steps until the performance of the neural network model reaches the preset conditions and obtain the target model.

[0114] Optionally, the above-mentioned preset conditions may be a loss function threshold, an accuracy threshold, a mean absolute error threshold, a determination coefficient threshold, the number of iterations, a convergence condition, etc. If the generalization ability of the model evaluated in the third step does not meet the above-mentioned preset conditions (for example, the MSE needs to be lower than a certain threshold), it is continuously optimized based on feedback until the performance indicators of the model meet the preset standards or conditions to obtain the target model.

[0115] Through the above technical solution, the risk of overfitting of the neural network model is reduced, so that the model can still maintain high performance when facing new and unseen data, thereby achieving the technical effect of improving the prediction accuracy of the battery health status.

[0116] In an optional embodiment, processing the state data to obtain the target data includes:

[0117] The first step is to obtain from the status data the voltage of the target battery at the end of charging, the voltage of the target battery within a preset time after the end of charging, and the current data collected from the target battery within a preset time interval.

[0118] For example, the voltage of the target battery at the end of charging is 4.0 V, and the voltage 0.1 second after the end of charging is 3.92 V. Assuming that current data is collected every 10 seconds, the data is collected for a total of 3600 seconds.

[0119] In the second step, the voltage drop of the target battery is obtained by subtracting the voltage of the target battery within a preset time after the charging is completed from the voltage of the target battery at the charging completion time.

[0120] For example, according to the numerical values ​​in the first step, the voltage drop of the target battery is calculated to be 4V-3.92V=0.08V.

[0121] The third step is to calculate the average current of the target battery during the discharge period based on the current data collected from the target battery within a preset time interval.

[0122] For example, the average current during the target battery discharge period can be calculated to be 20.5A from the 3600 seconds of current data collected in the first step.

[0123] In the fourth step, the voltage of the target battery at the end of charging, the voltage drop of the target battery, and the average current of the target battery during discharge are taken as target data.

[0124] For example, the target battery voltage drop of 0.08V, the average current of 20.5A during discharge, and the voltage of 4.0V at the end of charging are the target data, which are input into the target model for prediction processing and output the health status value of the target battery.

[0125] Through the above technical solution, the battery status data is processed to provide a data basis for the subsequent model to predict the battery health status value, thereby achieving the technical effect of real-time detection of the battery health status.

[0126] In an optional embodiment, the loss function is optimized using an automatic feature selection algorithm, and the optimized loss function includes:

[0127] In the first step, an automatic feature selection algorithm is used to calculate the contribution of the battery voltage at the end of charging, the battery voltage drop, and the average current during battery discharge in the training set to the prediction of the battery health status estimate, and multiple contribution values ​​are obtained.

[0128] Optionally, an automatic feature selection algorithm is used to measure the influence of the battery voltage at the end of charging, the battery voltage drop, and the average current during battery discharge on the model prediction result. The automatic feature selection algorithm may be L1 regularization, L2 regularization, etc.

[0129] For example, after LASSO regression processing, the contribution value of each feature in the training set to SOH prediction was obtained. The contribution value of the battery voltage at the end of charging was 0.12; the contribution value of the battery voltage drop was 0.54; and the contribution value of the average current during battery discharge was 0.25.

[0130] In the second step, the automatic feature selection algorithm is adjusted according to multiple contribution values ​​to obtain an adjusted automatic feature selection algorithm.

[0131] Optionally, based on the multiple contribution values ​​obtained in the first step, adjust the parameters of the automatic feature selection algorithm to optimize the feature selection process and ensure that the model can be trained based on the most effective features. This adjustment will make the model more inclined to reduce the weights of features with smaller contributions during training, thereby achieving automatic feature screening.

[0132] For example, in the first step, the initialization setting of the LASSO regression parameter is 0.01 (the larger the λ in LASSO regression, the more the model tends to select fewer features). Based on the contribution value obtained in the first step, it is found that the contribution value of the battery voltage at the end of charging is small. To ensure that the model only retains features that contribute greatly to the prediction of SOH, the value of λ can be increased, for example, λ can be adjusted to 0.1.

[0133] The third step is to optimize the loss function using the adjusted automatic feature selection algorithm to obtain the optimized loss function.

[0134] Optionally, based on the loss function, an adjusted automatic feature selection algorithm is introduced to penalize input features whose weights are greater than a certain threshold, thereby obtaining an optimized loss function. This optimized loss function encourages the model to reduce feature redundancy while maintaining prediction accuracy, thereby improving the model's interpretability and generalization capabilities.

[0135] For example, the loss function can be optimized as Loss = MSE + λ(|w1|+|w2|+|w3|).

[0136] Through the above technical solution, it is ensured that the model can learn from key features, achieving the technical effect of improving the model's operating speed and prediction accuracy.

[0137] In an optional embodiment, after the neural network model is trained using the training set to obtain the neural network model to be tested, the method further includes:

[0138] The first step is to use interpretability tools to perform interpretability analysis on the neural network model to be tested and obtain the interpretability analysis results.

[0139] Optionally, the report or data on the model decision process obtained by analyzing the model through an interpretability tool is the interpretability analysis result, which may include the contribution or importance score of each input feature. The above-mentioned interpretability tool refers to a software tool used to explain the decision process of the neural network, which may be SHAP, LIME, etc.

[0140] In the second step, based on the results of the interpretability analysis, the neural network model to be tested is adjusted until the interpretability index of the neural network model to be tested reaches the preset standard.

[0141] Optionally, based on the interpretability analysis results, structural adjustments or optimization algorithm adjustments may be made to the neural network model to be tested, until the interpretability index of the model (such as feature contribution distribution) reaches a preset standard.

[0142] For example, by analyzing the model with the SHAP tool, the interpretability analysis results (SHAP values ​​of each feature for the model's predicted SOH) are: the SHAP value of the battery voltage at the end of charging is 0.2, the SHAP value of the battery voltage drop is 0.5, and the SHAP value of the average current during battery discharge is 0.3. The above analysis results show that the battery voltage drop contributes the most to the prediction of the battery health status. Assuming that according to the experimental results, the absolute value of the SHAP value of the average current during battery discharge must reach 0.2, and the absolute value of the SHAP value of the voltage drop at the end of charging must be higher than 0.4, it means that the contribution distribution is consistent with the battery aging experimental data, and the model meets the standard in terms of interpretability. If a value is lower than the set standard, the model structure can be adjusted by adding hidden layers or neurons, or the algorithm can be optimized by adjusting the learning rate.

[0143] Through the above technical solution, the interpretability of the model is enhanced, making the decision-making process of the model transparent and understandable, thereby achieving the technical effect of improving the prediction accuracy of the model.

[0144] The method for estimating the state of health of a battery provided in an embodiment of the present application obtains state data of a target battery, wherein the state data at least includes the voltage of the target battery at the end of charging, the voltage of the target battery within a preset time after the end of charging, and the current data during the discharge period collected from the target battery within a preset time interval; processes the state data to obtain target data; and inputs the target data into a target model for predictive processing to obtain an estimated value of the state of health of the target battery, wherein the target model is an optimized neural network model obtained by training and optimizing a neural network model using a data set, the data set including multiple groups of battery test data, each group of test data including at least: the voltage of the battery at the end of charging, the voltage drop of the battery, the average current during the discharge of the battery, and the state of health value of the battery. The method solves the problem in the prior art that the evaluation of the state of health of a battery generally relies on the charge and discharge cycle data of homogeneous batteries for working condition lookup, resulting in low accuracy of the evaluation of the state of health of the battery. The target data obtained based on the state data of the target battery is input into the optimized neural network model, and the optimized neural network model is used to estimate the state of health of the battery, thereby achieving the technical effect of improving the accuracy of the evaluation of the state of health of the battery.

[0145] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0146] Example 2

[0147] An embodiment of the present application may provide an electronic device, Figure 6 is a structural block diagram of an electronic device according to an embodiment of the present application. Figure 6 As shown, the electronic device may include: one or more ( Figure 6 (only one is shown) processor 602, memory 604, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0148] Among them, the memory can be used to store software programs and modules, such as program instructions / modules corresponding to the methods and devices in the embodiments of the present application, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0149] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtain the status data of the target battery, wherein the status data at least includes the voltage of the target battery at the end of charging, the voltage of the target battery within a preset time after the end of charging, and the current data during the discharge period collected from the target battery within a preset time interval; process the status data to obtain target data; input the target data into the target model for predictive processing to obtain the health status estimate of the target battery, wherein the target model is an optimized neural network model obtained by training and optimizing the neural network model using a data set, and the data set includes multiple groups of battery test data, and each group of test data includes at least: the voltage of the battery at the end of charging, the voltage drop of the battery, the average current during the discharge of the battery, and the health status value of the battery.

[0150] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: perform a charging test on the battery according to a preset charging strategy, and after charging is completed, record the voltage of the battery at the end of charging and the voltage of the battery within a preset time after the end of charging; use the voltage of the battery at the end of charging minus the voltage of the battery within a preset time after the end of charging to obtain the voltage drop of the battery; perform a discharge test on the battery according to a preset discharge strategy, and during the discharge process, collect the current data of the battery according to preset time intervals to calculate the average current during the discharge of the battery; after the discharge is completed, record the discharge capacity of the battery, calculate the ratio of the discharge capacity to the initial capacity of the battery, and obtain the health status value of the battery; obtain each group of test data based on the voltage of the battery at the end of charging, the voltage drop of the battery, the average current during the discharge of the battery, and the health status value of the battery; repeat the above steps until the battery life cycle reaches the termination condition, obtain multiple groups of test data, and obtain a data set based on the multiple groups of test data.

[0151] The processor can call the information and application programs stored in the memory through the transmission device to perform the following steps: divide the data set into a training set and a test set; use the training set to train the neural network model to obtain the neural network model to be tested; use the test set to test the neural network model to be tested to obtain the target model.

[0152] The processor can call the information and application programs stored in the memory through the transmission device to perform the following steps: based on the data set, the voltage of the battery at the end of charging, the voltage drop of the battery, and the average current during the discharge of the battery are used as input features of the input layer of the neural network model; multiple hidden layers are designed in the neural network model, wherein each hidden layer includes at least a preset activation function and a preset number of neurons; and the health status value of the battery is used as the output target of the output layer of the neural network model.

[0153] The processor can call the information and application programs stored in the memory through the transmission device to perform the following steps: initialize the weights and bias parameters of the multi-layer hidden layers and output layers in the neural network model; input the voltage of the battery at the end of charging, the voltage drop of the battery, and the average current during the discharge of the battery in the training set into the neural network model, execute the forward propagation algorithm, and obtain the estimated value of the health state of the battery; use the automatic feature selection algorithm to optimize the loss function to obtain the optimized loss function; quantify the difference between the estimated value of the health state of the battery and the actual value of the health state of the battery in the training set according to the optimized loss function; based on the difference, use the back propagation algorithm to calculate the gradient of the optimized loss function relative to the weights and bias parameters of the multi-layer hidden layers and the output layer; use the optimization algorithm to update the weights and bias parameters of the multi-layer hidden layers and the output layer according to the gradient; repeat the above steps until the neural network model reaches the preset training stop condition.

[0154] The processor can call the information and application programs stored in the memory through the transmission device to perform the following steps: input the voltage of the battery in the test set at the end of charging, the voltage drop of the battery, and the average current during the battery discharge into the neural network model to be tested, and execute the forward propagation algorithm to obtain the estimated health state of the battery; calculate the preset performance index to quantify the difference between the estimated health state of the battery and the actual health state of the battery in the test set; evaluate the generalization ability of the neural network model based on the performance index; repeat the above steps until the performance of the neural network model reaches the preset conditions and the target model is obtained.

[0155] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtain the voltage of the target battery at the end of charging, the voltage of the target battery within a preset time after the end of charging, and the current data collected from the target battery within a preset time interval from the status data; subtract the voltage of the target battery within a preset time after the end of charging from the voltage of the target battery at the end of charging to obtain the voltage drop of the target battery; calculate the average current of the target battery during discharge based on the current data collected from the target battery within the preset time interval; use the voltage of the target battery at the end of charging, the voltage drop of the target battery, and the average current of the target battery during discharge as target data.

[0156] The processor can call the information and application programs stored in the memory through the transmission device to perform the following steps: use the automatic feature selection algorithm to calculate the contribution of the battery voltage at the end of charging, the battery voltage drop, and the average current during battery discharge in the training set to the prediction of the battery health status estimate, and obtain multiple contribution values; adjust the automatic feature selection algorithm according to the multiple contribution values ​​to obtain an adjusted automatic feature selection algorithm; optimize the loss function using the adjusted automatic feature selection algorithm to obtain an optimized loss function.

[0157] The processor can call the information and application programs stored in the memory through the transmission device to perform the following steps: use the interpretability tool to perform an interpretability analysis on the neural network model to be tested to obtain the interpretability analysis results; according to the interpretability analysis results, adjust the neural network model to be tested until the interpretability index of the neural network model to be tested reaches the preset standard.

[0158] By adopting the embodiment of the present application, a battery health state estimation scheme is provided. By acquiring the state data of the target battery, wherein the state data at least includes the voltage of the target battery at the end of charging, the voltage of the target battery within a preset time after the end of charging, and the current data during the discharge period collected from the target battery within a preset time interval; processing the state data to obtain the target data; inputting the target data into the target model for prediction processing to obtain the health state estimation value of the target battery, wherein the target model is an optimized neural network model obtained by training and optimizing the neural network model using a data set, and the data set includes multiple groups of battery test data, each group of test data includes at least: the voltage of the battery at the end of charging, the voltage drop of the battery, the average current during the battery discharge, and the health state value of the battery, the problem that the battery health state evaluation in the prior art generally relies on the charge and discharge cycle data of homogeneous batteries for working condition lookup, resulting in low accuracy of the battery health state evaluation, and using the neural network model to capture the nonlinear and complex relationship between the input features and the battery health state value, thereby achieving the technical effect of improving the accuracy of battery health state prediction.

[0159] It can be understood by those skilled in the art that Figure 6 The structure shown is for illustration only, and the electronic device may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (Mobile Internet Devices, MID), a PAD, or other terminal devices. Figure 6 The structure of the electronic device is not limited. Figure 6 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Figure 6Different configurations shown.

[0160] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

Claims

1. A method for estimating a battery health state, characterized in that: include: Acquire status data of the target battery, wherein the status data at least includes the voltage of the target battery at the end of charging, the voltage of the target battery within a preset time after the end of charging, and the current data of the target battery during discharge collected within a preset time interval; Processing the state data to obtain target data; The target data is input into a target model for prediction processing to obtain an estimated value of the health status of the target battery, wherein the target model is an optimized neural network model obtained by training and optimizing a neural network model using a data set, and the data set includes multiple groups of battery test data, and each group of test data includes at least: the voltage of the battery at the end of charging, the voltage drop of the battery, the average current during battery discharge, and the health status value of the battery.

2. The method according to claim 1, characterized in that The data set is obtained by the following steps: Perform a charging test on the battery according to a preset charging strategy, and after charging is completed, record the voltage of the battery at the end of charging and the voltage of the battery within a preset time after charging is completed; The voltage drop of the battery is obtained by subtracting the voltage of the battery within a preset time after the charging is completed from the voltage of the battery at the time when the charging is completed; Perform a discharge test on the battery according to a preset discharge strategy. During the discharge process, collect the battery current data according to a preset time interval and calculate the average current of the battery during discharge; After the discharge is completed, the discharge capacity of the battery is recorded, and the ratio of the discharge capacity to the initial capacity of the battery is calculated to obtain the health status value of the battery; Obtain each set of test data according to the voltage of the battery at the end of charging, the voltage drop of the battery, the average current during the discharge of the battery, and the health status of the battery; The above steps are repeatedly performed until the battery life cycle reaches a termination condition, and multiple groups of test data are obtained, and the data set is obtained based on the multiple groups of test data.

3. The method according to claim 2, characterized in that The target model is obtained by the following steps: Dividing the data set into a training set and a test set; Using the training set to train the neural network model to obtain a neural network model to be tested; The test set is used to test the neural network model to be tested to obtain the target model.

4. The method according to claim 3, characterized in that Before using the training set to train the neural network model, the method further includes: According to the data set, the voltage of the battery at the end of charging, the voltage drop of the battery, and the average current of the battery during discharge are used as input features of the input layer of the neural network model; Designing multiple hidden layers in the neural network model, wherein each hidden layer includes at least a preset activation function and a preset number of neurons; The health status value of the battery is used as the output target of the output layer of the neural network model.

5. The method according to claim 4, characterized in that Using the training set to train the neural network model includes: Initializing weights and bias parameters of the multiple hidden layers and the output layer in the neural network model; Inputting the voltage of the battery at the end of charging, the voltage drop of the battery, and the average current of the battery during discharge in the training set into the neural network model, executing a forward propagation algorithm, and obtaining an estimated value of the health state of the battery; The loss function is optimized using an automatic feature selection algorithm to obtain an optimized loss function; quantifying the difference between the estimated state of health of the battery and the actual state of health of the battery in the training set according to the optimized loss function; According to the difference, using a back propagation algorithm, calculate the gradient of the optimized loss function relative to the weights and bias parameters of the multi-layer hidden layer and the output layer; Using an optimization algorithm, updating the weights and bias parameters of the multiple hidden layers and the output layer according to the gradient; The above steps are repeated until the neural network model reaches a preset training stop condition.

6. The method according to claim 3, characterized in that The neural network model to be tested is tested using the test set to obtain a target model including: Inputting the voltage of the battery in the test set at the end of charging, the voltage drop of the battery, and the average current of the battery during discharge into the neural network model to be tested, executing a forward propagation algorithm, and obtaining an estimated value of the health state of the battery; Calculating a preset performance indicator to quantify the difference between the estimated state of health of the battery and the actual state of health of the battery in the test set; According to the performance index, evaluating the generalization ability of the neural network model; The above steps are repeated until the performance of the neural network model reaches a preset condition, thereby obtaining a target model.

7. The method according to claim 1, characterized in that Processing the state data to obtain target data includes: Acquire, from the status data, the voltage of the target battery at the end of charging, the voltage of the target battery within a preset time after the end of charging, and the current data collected from the target battery within a preset time interval; The voltage drop of the target battery is obtained by subtracting the voltage of the target battery within a preset time after the charging is completed from the voltage of the target battery at the charging completion time. Calculating an average current of the target battery during discharge according to the current data collected from the target battery within the preset time interval; The voltage of the target battery at the end of charging, the voltage drop of the target battery, and the average current of the target battery during discharge are used as the target data.

8. The method according to claim 5, characterized in that The automatic feature selection algorithm is used to optimize the loss function, and the optimized loss function includes: Using an automatic feature selection algorithm, calculating the contribution of the voltage of the battery at the end of charging, the voltage drop of the battery, and the average current of the battery during discharge in the training set to the prediction of the estimated value of the battery health state, respectively, to obtain a plurality of contribution values; Adjusting the automatic feature selection algorithm according to the multiple contribution values ​​to obtain an adjusted automatic feature selection algorithm; The adjusted automatic feature selection algorithm is used to optimize the loss function to obtain the optimized loss function.

9. The method according to claim 3, characterized in that: After the neural network model is trained using the training set to obtain a neural network model to be tested, the method further includes: Using an interpretability tool to perform an interpretability analysis on the neural network model to be tested to obtain an interpretability analysis result; According to the interpretability analysis result, the neural network model to be tested is adjusted until the interpretability index of the neural network model to be tested reaches a preset standard.

10. An electronic device, characterized in that: include: A memory storing an executable program; A processor, configured to run the program, wherein the program, when running, executes the method for estimating the battery health status as described in any one of claims 1 to 9.

Images