Lithium battery charging state determination method and apparatus, and electronic device

Abstract

The invention discloses a lithium battery charging state determination method and apparatus, and an electronic device. The method comprises the steps that a target model is obtained, and the target model is obtained by conducting pre-training based on a first electrochemical impedance spectroscopy data set and conducting retraining based on a second electrochemical impedance spectroscopy data set, the first electrochemical impedance spectroscopy data set and the second electrochemical impedance spectroscopy data set are sets of electrochemical impedance spectroscopy data of a first type of lithium battery and a second type of lithium battery under excitation current with preset frequency respectively; acquiring target electrochemical impedance spectroscopy data of a second type of target lithium battery under the excitation current with the predetermined frequency; and inputting the target electrochemical impedance spectroscopy data into the target model to obtain the charging state of the target lithium battery. According to the invention, the technical problem that the efficiency is low when the charging states of different types of lithium batteries are determined according to the electrochemical impedance spectroscopy data due to high collection cost of the electrochemical impedance spectroscopy data in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of batteries, and in particular to a method and device for determining the charging state of a lithium battery, and electronic equipment. Background Art

[0002] The state of charge (SOC) of a lithium battery is an important indicator for evaluating lithium batteries. In related technologies, the coulomb counting method relies on a complete charging and discharging process and cannot meet the needs of real-time SOC estimation. Model-based estimation, such as equivalent circuit models and Kalman filters, simplifies the complexity of the battery, but its accuracy decreases significantly with the dynamic evolution of battery conditions such as aging, temperature changes, and lithium deposition. Although electrochemical models improve modeling accuracy, their application in real-time SOC estimation is limited due to the difficulty in obtaining chemical parameters and the complexity of partial differential equation processing. Electrochemical impedance spectroscopy (EIS) measurement can provide information on the dynamic characteristics of the battery, but data collection is costly and time-consuming.

[0003] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0004] Embodiments of the present invention provide a method, device, and electronic device for determining the charging state of a lithium battery, to at least address the technical problem in the related art of high cost of collecting electrochemical impedance spectroscopy data, resulting in low efficiency in determining the charging state of different types of lithium batteries based on electrochemical impedance spectroscopy data.

[0005] According to one aspect of an embodiment of the present invention, a method for determining a charging state of a lithium battery is provided, comprising: obtaining a target model, wherein the target model is pre-trained based on a first electrochemical impedance spectroscopy data set and retrained based on a second electrochemical impedance spectroscopy data set, the first electrochemical impedance spectroscopy data set being a set of electrochemical impedance spectroscopy data of a first type of lithium battery under an excitation current of a predetermined frequency, and the second electrochemical impedance spectroscopy data set being a set of electrochemical impedance spectroscopy data of a second type of lithium battery under an excitation current of the predetermined frequency; obtaining target electrochemical impedance spectroscopy data of a target lithium battery of the second type under the excitation current of the predetermined frequency; and inputting the target electrochemical impedance spectroscopy data into the target model to obtain the charging state of the target lithium battery.

[0006] Optionally, the first electrochemical impedance spectroscopy data set is obtained based on the following method: charging the first type of lithium battery in a fully discharged state; when multiple charging states are reached, the first type of lithium battery is allowed to stand for a predetermined time; and applying an excitation current of the predetermined frequency to the first type of lithium battery after standing, to obtain the first electrochemical impedance spectroscopy data set when multiple charging states are reached.

[0007] Optionally, the second electrochemical impedance spectroscopy data set is obtained based on the following manner: charging the second type of lithium battery in a fully discharged state; when multiple charging states are reached, applying the excitation current of the predetermined frequency to the second type of lithium battery respectively, and obtaining the second electrochemical impedance spectroscopy data set when multiple charging states are reached.

[0008] Optionally, the target model is obtained based on the following training method, including: constructing an initial model; pre-training the initial model based on the first electrochemical impedance spectroscopy data set, determining the weights and bias values ​​of multiple hidden layers of the initial model, and obtaining an intermediate model, wherein the weights and bias values ​​of the multiple hidden layers are used to characterize the first relationship between the electrochemical impedance spectroscopy data and the charging state of the first type of lithium battery; retraining the intermediate model based on the second electrochemical impedance spectroscopy data set and part of the first electrochemical impedance spectroscopy data set, adjusting the weights and bias values ​​of the last hidden layer of the intermediate model, and obtaining the target model, wherein the weights and bias values ​​of the other hidden layers in the multiple hidden layers except the last hidden layer and the adjusted last hidden layer are used to characterize the second relationship between the electrochemical impedance spectroscopy data and the charging state of the second type of lithium battery, and the adjustment of the weights and bias of the last hidden layer is used to characterize the difference between the first relationship and the second relationship.

[0009] Optionally, the first electrochemical impedance spectroscopy data set includes a normalized first resistance data set and a first reactance data set, and the second electrochemical impedance spectroscopy data set includes a normalized second resistance data set and a second reactance data set.

[0010] Optionally, a difference between the number of data in the first electrochemical impedance spectroscopy data set and the number of data in the second electrochemical impedance spectroscopy data set is greater than a predetermined threshold.

[0011] According to another aspect of the present invention, a device for determining a charging state of a lithium battery is provided, comprising: a first acquisition module for acquiring a target model, wherein the target model is pre-trained based on a first electrochemical impedance spectroscopy data set and retrained based on a second electrochemical impedance spectroscopy data set, the first electrochemical impedance spectroscopy data set being a set of electrochemical impedance spectroscopy data of a first type of lithium battery under an excitation current of a predetermined frequency, and the second electrochemical impedance spectroscopy data set being a set of electrochemical impedance spectroscopy data of a second type of lithium battery under an excitation current of the predetermined frequency; a second acquisition module for acquiring target electrochemical impedance spectroscopy data of a target lithium battery of the second type under the excitation current of the predetermined frequency; and an input module for inputting the target electrochemical impedance spectroscopy data into the target model to obtain the charging state of the target lithium battery.

[0012] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned methods for determining the charging status of a lithium battery.

[0013] According to another aspect of the present invention, an electronic device is provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the program executes any one of the above-mentioned methods for determining the charging status of a lithium battery when running.

[0014] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the steps of any one of the methods for determining a charging state of a lithium battery are implemented.

[0015] In an embodiment of the present invention, a target model is obtained, wherein the target model is pre-trained based on a first electrochemical impedance spectroscopy data set and retrained based on a second electrochemical impedance spectroscopy data set, the first electrochemical impedance spectroscopy data set being a set of electrochemical impedance spectroscopy data of a first type of lithium battery under an excitation current of a predetermined frequency, and the second electrochemical impedance spectroscopy data set being a set of electrochemical impedance spectroscopy data of a second type of lithium battery under an excitation current of a predetermined frequency; target electrochemical impedance spectroscopy data of a second type of target lithium battery under an excitation current of a predetermined frequency is obtained; and the target electrochemical impedance spectroscopy data is input into the target model to obtain a charge state of the target lithium battery. This achieves the purpose of obtaining a charge state of the second type of target lithium battery through transfer learning based on the electrochemical impedance spectroscopy data sets of the first type of lithium battery and the second type of lithium battery, thereby achieving the technical effect of reducing the amount of electrochemical impedance spectroscopy data collected and improving the efficiency of determining the charge state of the lithium battery, thereby solving the technical problem in the related art that the cost of collecting electrochemical impedance spectroscopy data is high and the efficiency of determining the charge state of different types of lithium batteries based on electrochemical impedance spectroscopy data is low. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0017] Figure 1 is a flow chart of a method for determining a charging state of a lithium battery according to an embodiment of the present invention;

[0018] Figure 2 is a schematic diagram of an SOC prediction architecture based on transfer learning according to an optional embodiment of the present invention;

[0019] Figure 3 is a schematic diagram showing a comparison of SOC prediction accuracy of different models according to an optional embodiment of the present invention;

[0020] Figure 4 4 is a structural block diagram of a device for determining a charging state of a lithium battery according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention 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 numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes 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.

[0023] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following interpretations:

[0024] State of Charge (SOC), also known as state of charge, is a key parameter in battery technology that quantifies the remaining available energy in a battery. SOC is defined as the ratio of the remaining battery capacity to the battery's maximum capacity and can be expressed as a percentage (0% to 100%). Accurately estimating SOC not only reflects the battery's current state but also improves battery reliability by preventing overcharging and overdischarging, making it crucial for ensuring efficient and safe battery operation.

[0025] Electrochemical impedance spectroscopy (EIS) is a testing technique used to analyze the frequency-dependent impedance changes of electrochemical systems, such as batteries, electrodes, and electrolytes. By applying a small AC signal of known frequency and amplitude and measuring the amplitude and phase of the system response, EIS can determine the impedance changes of the electrochemical system. The amplitude of the AC signal is typically in the millivolt range, and the frequency range can range from a few hertz to tens of kilohertz or even higher. EIS can provide in-depth information about the internal dynamics, interfacial reactions, and diffusion processes of the battery, which is crucial for understanding battery performance, health status, and life prediction. Its operating principle is based on the frequency response analysis of the impedance of the electrochemical system. EIS data can be displayed in the form of Nyquist and Bode plots. The Nyquist plot shows the relationship between the real and imaginary parts of the complex impedance, while the Bode plot plots the impedance modulus and phase as a function of frequency.

[0026] Lithium iron phosphate (LFP) batteries are a type of lithium battery. Their cathode material is a lithium iron phosphate compound composed of lithium (Li), iron (Fe), phosphorus (P), and oxygen (O), with the molecular formula LiFePO4. This unique combination of materials gives LFP batteries high safety, long life, excellent cost-effectiveness, and environmental friendliness.

[0027] Nickel Cobalt Manganese Lithium Oxide (NCM) batteries are also known as lithium batteries. The positive electrode material of NCM batteries is a composite of three metal oxides: nickel (Ni), cobalt (Co), and manganese (Mn). The ratio can be adjusted according to specific application requirements. Common formulations include NCM111 (equimolar ratio), NCM523, NCM622, and NCM811. NCM batteries offer high energy density, excellent cycle performance, and wide applicability.

[0028] The state of charge (SOC) of a lithium battery is an important metric for evaluating it. SOC estimation methods in related technologies all have limitations. For example, the coulomb counting method can be used to calculate SOC, estimating SOC by integrating the current flowing through the battery during charging or discharging. However, this method requires a relatively complete charge / discharge cycle to determine the battery capacity, limiting its application in scenarios requiring real-time SOC estimation. Mathematical modeling methods can also be used to calculate SOC. Various controllers (such as Kalman filters and particle filters) update the measured values ​​of the lithium battery and estimate SOC based on an equivalent circuit model (ECM) of the lithium battery, which can be represented by a resistor-capacitor circuit. However, this equivalent circuit model oversimplifies the lithium battery, and its accuracy decreases significantly when the battery state changes (such as battery aging, temperature changes, and lithium deposition). Because the accuracy of model-based estimators is highly dependent on the accuracy of the battery model, electrochemical models (EM) have been proposed as an alternative method for high-precision modeling of lithium batteries. Although electrochemical models improve the accuracy of battery modeling, the difficulty in obtaining chemically related parameters and handling partial differential equations limits their potential for real-time SOC estimation in practical applications.

[0029] Given the limitations of the aforementioned methods, data-driven SOC estimation techniques have been proposed in the related art. EIS measurements provide extensive information about battery dynamics across a wide frequency range, but in practice, acquiring sufficient data to train new models is time-consuming and expensive. The distribution of EIS data collected at different rest times varies due to the varying chemical states within the battery. For different lithium-ion battery types, the distribution of collected EIS data varies due to differences in battery chemistry, capacity, and structure. Data-driven methods in the related art assume that training and test data come from the same experimental source and have similar distributions. This assumption means that a statistical model trained on one dataset may not be seamlessly applicable to another dataset with a different distribution. In such cases, the model must be rebuilt and retrained from scratch using a new dataset. However, in real-world scenarios, acquiring sufficient data to retrain a new model can be a time-consuming and expensive process, and the training phase itself requires significant resources. Due to the effects of rest on internal impedance and the diverse material designs in lithium-ion batteries, EIS data collected in practice has varying distributions, making it difficult to achieve effective results in SOC estimation using real-time EIS data.

[0030] In light of this, embodiments of the present invention propose transfer learning as a solution. This approach leverages knowledge gained from a source task (i.e., determining the state of charge of a first type of lithium battery) to address a different but related target task (i.e., determining the state of charge of a second type of lithium battery). This approach achieves high determination efficiency and accuracy, even with potentially different data distributions. This approach makes model retraining more efficient, requiring less data.

[0031] According to an embodiment of the present invention, an embodiment of a method for determining the charging state of a lithium battery is 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.

[0032] Figure 1 FIG. 1 is a flow chart of a method for determining a charging state of a lithium battery according to an embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0033] Step S102, obtaining a target model, wherein the target model is pre-trained based on a first electrochemical impedance spectroscopy data set and retrained based on a second electrochemical impedance spectroscopy data set, the first electrochemical impedance spectroscopy data set is a set of electrochemical impedance spectroscopy data of a first type of lithium battery under an excitation current of a predetermined frequency, and the second electrochemical impedance spectroscopy data set is a set of electrochemical impedance spectroscopy data of a second type of lithium battery under an excitation current of a predetermined frequency.

[0034] As an optional embodiment, the execution subject of the method of this embodiment can be a terminal or server used to determine the charging status of a lithium battery, utilizing the advanced processing capabilities and storage resources of the terminal or server to train and use the target model. The execution subject of the method of this embodiment can also be the battery management system where the target lithium battery is located, that is, a module with the function of determining the charging status of a lithium battery is deeply embedded in the battery management system where the target lithium battery is located, realizing seamless and intelligent determination of the charging status of the lithium battery, and broadening the scope of application and scenario adaptability of the method.

[0035] As an optional embodiment, the target model is the key to realizing the method for determining the charging state of a lithium battery, and it is realized through two stages of data training. In the first stage, the model is pre-trained based on a first electrochemical impedance spectroscopy data set from a first type of lithium battery (such as an LFP battery) to capture the relationship between the electrochemical characteristics and the charging state of the first type of battery at a specific excitation current frequency, and to construct a basic model framework that initially has the ability to determine the charging state of a lithium battery. In the second stage, the model is further retrained based on a second electrochemical impedance spectroscopy data set from a second type of lithium battery (such as an NCM battery) to adjust the model to adapt to the relationship between the electrochemical characteristics and the charging state of the second type of battery at a specific excitation current frequency. The target model obtained by the above training method has a higher generalization ability, improves the accuracy and applicability of the target model in determining the charging state of different types of lithium batteries, and can also significantly reduce the amount of electrochemical impedance spectroscopy data collected corresponding to the second type of lithium battery, thereby reducing data collection costs and time investment.

[0036] As an optional embodiment, the first electrochemical impedance spectroscopy data set can be obtained based on a variety of methods, for example, charging the first type of lithium battery in a fully discharged state; when multiple charging states are reached, the first type of lithium battery is allowed to stand for a predetermined time; an excitation current of a predetermined frequency is applied to the first type of lithium battery after standing, and the first electrochemical impedance spectroscopy data set is obtained when multiple charging states are reached. By placing the first type of lithium battery in a fully discharged state and then charging it until multiple predetermined charging states are reached, it can be ensured that the first electrochemical impedance spectroscopy data set obtained can cover all charging states of the lithium battery. By allowing the first type of lithium battery to stand for a predetermined time (e.g., 1 hour), the lithium battery can be controlled to be in a stable state, thereby ensuring the quality and consistency of the electrochemical impedance spectroscopy data. The first electrochemical impedance spectroscopy data set obtained by the above method can characterize the electrochemical characteristics of the first type of lithium battery when it reaches stability under multiple charging states, avoid being affected by electrochemical reactions when it is unstable, and better reflect the electrochemical characteristics of the first type of lithium battery itself.

[0037] As an optional embodiment, the second electrochemical impedance spectroscopy data set can be obtained based on a variety of methods, for example, charging the second type of lithium battery in a fully discharged state; when reaching multiple charging states, applying an excitation current of a predetermined frequency to the second type of lithium battery respectively, and obtaining the second electrochemical impedance spectroscopy data set when reaching multiple charging states. By placing the second type of lithium battery in a fully discharged state and then charging it until reaching multiple predetermined charging states, it can be ensured that the obtained second electrochemical impedance spectroscopy data set can cover all charging states of the lithium battery. Unlike the first electrochemical impedance spectroscopy data set, when obtaining the second electrochemical impedance spectroscopy data set, the second type of lithium battery can be left to stand for a predetermined time. Instead, the excitation current is directly applied after reaching multiple predetermined charging states, and the corresponding electrochemical impedance spectroscopy data are collected respectively. The second electrochemical impedance spectroscopy data set obtained by the above method can not only characterize the dynamic characteristics of the second type of lithium battery in multiple charging states, but also reduce the collection time of the second electrochemical impedance spectroscopy data set by avoiding standing.

[0038] As an optional embodiment, the target model can be obtained in multiple ways based on the following training method, for example, constructing an initial model; pre-training the initial model based on the first electrochemical impedance spectroscopy data set, determining the weights and bias values ​​of multiple hidden layers of the initial model, and obtaining an intermediate model, wherein the weights and bias values ​​of the multiple hidden layers are used to characterize the first relationship between the electrochemical impedance spectroscopy data and the charging state of the first type of lithium battery; retraining the intermediate model based on the second electrochemical impedance spectroscopy data set and part of the first electrochemical impedance spectroscopy data set, adjusting the weights and bias values ​​of the last hidden layer of the intermediate model, and obtaining a target model, wherein the weights and bias values ​​of the other hidden layers except the last hidden layer in the multiple hidden layers and the adjusted last hidden layer are used to characterize the second relationship between the electrochemical impedance spectroscopy data and the charging state of the second type of lithium battery, and the adjustment of the weights and bias of the last hidden layer is used to characterize the difference between the first relationship and the second relationship. Through the above training method, for different second types of lithium batteries, only the weights and bias values ​​of the last hidden layer of the model can be adjusted, thereby improving the role of pre-training, making full use of the features learned in pre-training, and avoiding repeated learning of the common parts of the electrochemical characteristics of lithium batteries, thereby obtaining the target model more quickly.

[0039] As an optional embodiment, the first electrochemical impedance spectroscopy data set includes a normalized first resistance data set and a first reactance data set, and the second electrochemical impedance spectroscopy data set includes a normalized second resistance data set and a second reactance data set. Based on the complex electrochemical impedance spectroscopy data set, a real-part resistance data set and an imaginary-part reactance data set can be obtained, thereby enriching the model training set and providing more effective information. By normalizing the data set, it is possible to ensure that the data input to the model is on a uniform scale, avoiding model training bias caused by differences in data magnitude, and further improving the model's learning efficiency and prediction accuracy.

[0040] As an optional embodiment, the difference between the number of data in the first electrochemical impedance spectroscopy data set and the number of data in the second electrochemical impedance spectroscopy data set is greater than a predetermined threshold. A large amount of historical data from the first type of lithium battery can enhance the model's basic learning capabilities, while a small amount of historical data from the second type of lithium battery can fine-tune the model's prediction direction, achieving an optimal balance between data utilization efficiency and prediction accuracy. In practice, this strategy greatly reduces data acquisition costs and accelerates the model's adaptation to new battery types, making real-time, accurate estimation of lithium battery state of charge more feasible and efficient.

[0041] Step S104 : acquiring target electrochemical impedance spectroscopy data of a second type of target lithium battery under an excitation current of a predetermined frequency.

[0042] As an optional embodiment, the target lithium battery may be the lithium battery from which the above-mentioned second electrochemical impedance spectrum data set is collected, or it may be another battery of the same type as the lithium battery from which the above-mentioned second electrochemical impedance spectrum data set is collected. When there are multiple target lithium batteries, and they correspond to multiple types respectively, the electrochemical impedance spectrum data sets corresponding to the multiple types of lithium batteries may be used respectively, and the model may be retrained based on the multiple sets respectively. There are multiple ways to obtain the target electrochemical impedance spectrum data of the second type of target lithium battery under an excitation current of a predetermined frequency. For example, by applying a small-amplitude sinusoidal excitation current to the target lithium battery, the impedance response of the battery at a specific frequency is measured, i.e., the target electrochemical impedance spectrum data. The target electrochemical impedance spectrum data not only reflects the instantaneous electrochemical state of the target lithium battery, but also contains the dynamic information inside the target lithium battery. By obtaining the target electrochemical impedance spectrum data of the second type of target lithium battery under an excitation current of a predetermined frequency, the accuracy and real-time performance of the target model input data are ensured, which is the basis for determining the accuracy of the charging state of the lithium battery.

[0043] Step S106 , inputting the target electrochemical impedance spectroscopy data into the target model to obtain the charging state of the target lithium battery.

[0044] As an optional embodiment, the target model learns the relationship between electrochemical impedance spectroscopy data and the state of charge. Therefore, by inputting the target electrochemical impedance spectroscopy data into the target model, the state of charge of the target lithium battery can be obtained. Since the target model is pre-trained based on the first electrochemical impedance spectroscopy data set and retrained based on the second electrochemical impedance spectroscopy data set, the first electrochemical impedance spectroscopy data set is a set of electrochemical impedance spectroscopy data of the first type of lithium battery under an excitation current of a predetermined frequency, and the second electrochemical impedance spectroscopy data set is a set of electrochemical impedance spectroscopy data of the second type of lithium battery under an excitation current of a predetermined frequency, the target model fully learns the commonalities of the first type of lithium battery and the second type of lithium battery, and through retraining, it specifically learns the characteristics of the second type of lithium battery. Therefore, the state of charge of the target lithium battery can be obtained through the target model, overcoming the limitations of the relevant technology in processing different types of batteries.

[0045] Through the above steps, it is possible to obtain a target model, wherein the target model is pre-trained based on a first electrochemical impedance spectroscopy data set and retrained based on a second electrochemical impedance spectroscopy data set, the first electrochemical impedance spectroscopy data set being a set of electrochemical impedance spectroscopy data of a first type of lithium battery under an excitation current of a predetermined frequency, and the second electrochemical impedance spectroscopy data set being a set of electrochemical impedance spectroscopy data of a second type of lithium battery under an excitation current of a predetermined frequency; obtaining target electrochemical impedance spectroscopy data of a second type of target lithium battery under an excitation current of a predetermined frequency; inputting the target electrochemical impedance spectroscopy data into the target model to obtain a charge state of the target lithium battery, thereby achieving the purpose of obtaining a charge state of the second type of target lithium battery through transfer learning based on the electrochemical impedance spectroscopy data sets of the first type of lithium battery and the second type of lithium battery, thereby achieving the technical effect of reducing the amount of electrochemical impedance spectroscopy data collected and improving the efficiency of determining the charge state of the lithium battery, thereby solving the technical problem in the related art that the cost of collecting electrochemical impedance spectroscopy data is high and the efficiency of determining the charge state of different types of lithium batteries based on electrochemical impedance spectroscopy data is low.

[0046] In combination with the above-mentioned embodiments and optional embodiments, an optional implementation manner is provided. In this optional implementation manner, a method for overcoming the adaptability of the data-driven SOC prediction model to battery cells of different materials is proposed. This method solves the inherent limitations of data-driven SOC estimation, which stem from traditional assumptions about data distribution in machine learning technology. The differences in chemical materials, battery capacity and structure of lithium-ion batteries hinder the possibility of directly applying the SOC estimator developed for LFP batteries to the SOC estimation of NCM batteries. To overcome this challenge, in this method, transfer learning is used to combine the pre-trained SOC estimator of LFP batteries, so that the SOC estimation of NCM batteries can be achieved using a small amount of EIS measurement data collected from NCM batteries.

[0047] Figure 2 is a schematic diagram of an SOC prediction architecture based on transfer learning according to an optional embodiment of the present invention, such as Figure 2 As shown in Figure 1, the framework consists of two stages. In the first stage, a deep neural network (DNN) model is trained using the source dataset to estimate the SOC of the lithium battery based on the EIS measurement data specified in the source dataset. However, when the data collection method changes, the DNN model trained based on the source dataset may not be able to accurately estimate the SOC. This leads to the second stage, which is to accurately estimate the SOC of the lithium battery under different data collection scenarios (for example, for different types of lithium batteries).

[0048] The method includes the following processing.

[0049] S1, feature extraction.

[0050] Before inputting data into a machine learning model, data preprocessing plays a crucial role. Important data preprocessing techniques include data normalization and standardization. In this method, the input features of the EIS measurement data are normalized using a Min-Max function. After this process, the values ​​of each feature column are scaled to a range between 0 and 1. This normalization and scaling method ensures that each feature in the dataset has a balanced influence in the machine learning model, promoting model balance. Normalization not only helps improve the generalizability of the data but also retains all relevant information within a specified range, laying the foundation for robust and unbiased model training, thereby supporting meaningful analysis and accurate predictions. Each frequency measurement point corresponds to a complex impedance data point. Each EIS acquisition sample consists of 60 frequency measurement points, covering a range from high to low frequencies, to comprehensively reflect the dynamic characteristics of the battery at different excitation frequencies. Each complex impedance data point consists of a real part (Re(Z)) and an imaginary part (Im(Z)), representing the resistive and capacitive responses at that frequency, respectively. Therefore, the input features in a specific cycle contain 120 different values, which are defined as EIS features. The output corresponding to the input features is the capacity level at the measurement point. Feature extraction and statistical analysis of these complex impedance data can further reveal the electrochemical reaction process within the battery and provide high-dimensional feature information for subsequent battery state identification.

[0051] S2, source task and target task.

[0052] It should be noted that the source task refers to the initial goal during pre-training of the DNN model, while the target task is the specific scenario in which the retrained DNN model developed through transfer learning is applied.

[0053] In order to study the effects of battery rest and battery type on EIS measurements separately, an optional embodiment of the present invention includes two main stages. The first stage aims to optimize the effect of battery rest on EIS measurements through transfer learning. The goal is to promote the real-time application of EIS measurement data in SOC estimation of LFP batteries to cope with the limitations of EIS data collection in laboratory environments. The second stage aims to address the limitations of data-driven SOC estimation, which stem from common assumptions about data distribution in machine learning techniques. The changes in chemical materials, battery capacity and structure of lithium-ion batteries hinder the possibility of directly applying SOC estimators developed based on LFP batteries to SOC estimation of NCM batteries. To overcome this challenge, the second stage uses a transfer learning method in combination with the pre-trained SOC estimator of LFP batteries to achieve SOC estimation of NCM batteries using a small amount of EIS measurement data collected from NCM batteries.

[0054] Table 1 shows the horizontal relationship between each source task and its corresponding target task. In the first stage, the DNN model is pre-trained using EIS measurement data obtained from the steady state of the LFP battery, that is, after being charged and resting for a predetermined time. The target task is to estimate the SOC using real-time EIS measurement data obtained immediately after the LFP battery is charged, that is, without being rested after charging. In the second stage, the source task is to pre-train the DNN model using the EIS measurement data of the LFP battery, where the EIS measurement data of the LFP battery refers to the EIS measurement data after being rested. The target task is to estimate the SOC of the NCM battery, which is achieved by retraining the DNN model pre-trained in the source task using limited EIS measurement data for the NCM battery, where the limited EIS measurement data of the NCM battery refers to the EIS measurement data without being rested.

[0055] Table 1 Source tasks and target tasks

[0056]

[0057] S3, Research on the versatility of DNN-TL model based on transfer learning (TL) in NCM battery SOC estimation.

[0058] This method proposes a transfer learning-based DNN model, namely the DNN-TL model, to support data-driven SOC estimation for different types of lithium-ion batteries. To evaluate the performance of the proposed DNN-TL model on lithium-ion batteries with different rated capacities, three other different DNN-based SOC estimators were used to compare with the DNN-TL model. The model performance was evaluated using multiple regression metrics, including mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE), and coefficient of determination (R-squared, R2). Figure 3 is a schematic diagram comparing the SOC prediction accuracy of different models according to an optional embodiment of the present invention, such as Figure 3 shown.

[0059] The DNN-TL model is trained using EIS measurement data collected from LFP batteries at rest to capture the relationship between the electrochemical properties of LFP batteries and their capacity. This knowledge is stored in the weights and bias values ​​in the hidden layers of the DNN model. Subsequently, the pre-trained DNN model is retrained using EIS measurement data collected from NCM batteries at non-rest. During the retraining process, only the last hidden layer is fine-tuned, while the weights and biases of the other hidden layers remain unchanged. Through this retraining process, the knowledge from the EIS measurement data of LFP and NCM batteries is combined and stored in different hidden layers. Ultimately, the retrained DNN-TL model is able to reflect the chemical changes of NCM batteries by learning the commonalities between the two battery types. By transferring the knowledge from LFP batteries to NCM batteries, the DNN-TL model can estimate the SOC of NCM batteries using non-rest EIS measurement data. The experimental results show that the MSE, MAE, R-squared and MAPE of the DNN-TL model are 0.0122, 0.0706, 0.8940 and 0.1338 respectively.

[0060] The DNN-based SOC estimator 1 is a standalone DNN model with randomly initialized weights and biases, trained only using non-stationary EIS measurement data from NCM cells. During training, the weights and biases are optimized to capture the relationship between the electrochemical properties of NCM cells and their capacity. The trained standalone DNN model achieves SOC estimation performance of 0.0681 MSE, 0.1522 MAE, 0.4084 R-squared, and 0.3882. Because the standalone DNN model does not incorporate knowledge learned from LFP cells, it performs poorly with limited target data. In contrast, the DNN-TL model, retrained using the same amount of target data, provides more accurate SOC estimation for NCM cells. Furthermore, the target dataset is only approximately 20% of the source dataset collected from LFP cells, a significant reduction in size that is beneficial in real-world applications, where collecting EIS measurement data is often challenging.

[0061] DNN-based SOC estimator 2 is a DNN model trained using EIS measurement data from LFP batteries. This model is trained using EIS data collected from LFP batteries at rest and then uses EIS data collected from NCM batteries at non-rest to estimate SOC. This means that the SOC estimation model is trained based on the resting EIS data of LFP batteries and used to predict the SOC of NCM batteries, relying solely on non-resting EIS data. While data-driven approaches assume the same distribution for training and test datasets, this model exhibits limited generalization when battery types vary, and performs poorly when the test data distribution differs from the training data distribution. The evaluation metrics for SOC estimation using the DNN model trained on the resting EIS data of LFP batteries are as follows: MSE of 0.2033, MAE of 0.3624, R-squared of -1.0728, and MAPE of 0.3292. These results indicate that EIS data collected from LFP batteries cannot be directly used to estimate the SOC of NCM batteries due to differences in battery chemistry and capacity. Trained DNN models struggle to capture the dynamic chemical changes within batteries, limiting their scope of application. In contrast, the DNN-TL model overcomes the limitations of DNN models trained on static EIS measurement data from LFP batteries through a retraining process. By leveraging transfer learning, prior knowledge is leveraged to significantly improve the SOC estimation accuracy of non-static EIS measurement data from NCM batteries. It is worth noting that in machine learning, if the model's prediction performance is inferior to a simple strategy of predicting the mean of the target variable, the R-squared value may be negative. This indicates that the model fails to capture the inherent patterns of the data and its predictions are equivalent to simply predicting the mean.

[0062] DNN-based SOC estimator 3 is a DNN model trained on mixed EIS measurement data. It is trained on mixed EIS measurement data collected from LFP and NCM batteries. The trained DNN model is then used to estimate the SOC of NCM batteries, using only EIS measurement data collected from NCM batteries in a non-stationary state. The fusion of mixed datasets introduces changes in the distribution and diversity of EIS measurement data. Due to the significant difference in capacity between 52Ah LFP batteries and 3.6Ah NCM batteries, combining the datasets becomes complex and impractical. Furthermore, the smaller internal impedance of large-capacity batteries makes it difficult to fuse EIS measurement data from different batteries to support SOC estimation. The model failed to successfully estimate the SOC of NCM batteries using non-stationary measurement data. The differences in battery EIS measurement data also limit the model's applicability in real-world scenarios. In the DNN-TL model, EIS measurement data from different batteries are introduced into the DNN model at different training stages. In this model, however, the combination of mixed EIS measurement data leads to SOC estimation failure. Due to the extremely low estimation accuracy, the model performance cannot be reflected through evaluation indicators (MSE, MAE, R-squared and MAPE).

[0063] To improve the applicability of data-driven SOC estimators and ensure that the SOC estimator developed for LFP batteries can be extended to NCM batteries, a deep neural network based on transfer learning was employed to enable SOC estimation for NCM batteries using non-stationary EIS measurement data. The results show a significant improvement in SOC estimation accuracy, with a reduction of up to 82.08% in the mean error (MSE) and 53.15% in the mean error (MAE) compared to a standalone DNN model trained solely on NCM battery EIS measurement data. Furthermore, the data collection workload was significantly reduced, with the newly collected NCM battery EIS measurement data occupying only 20% of the source data size.

[0064] According to an embodiment of the present invention, a device for determining a charging state of a lithium battery is provided. Figure 4 is a structural block diagram of a device for determining a lithium battery charging state according to an embodiment of the present invention. Figure 4 As shown, the device includes: a first acquisition module 402, a second acquisition module 404 and an input module 406. The device is described below.

[0065] A first acquisition module 402 is used to acquire a target model, wherein the target model is pre-trained based on a first electrochemical impedance spectroscopy data set and retrained based on a second electrochemical impedance spectroscopy data set, the first electrochemical impedance spectroscopy data set is a set of electrochemical impedance spectroscopy data of a first type of lithium battery under an excitation current of a predetermined frequency, and the second electrochemical impedance spectroscopy data set is a set of electrochemical impedance spectroscopy data of a second type of lithium battery under an excitation current of a predetermined frequency; a second acquisition module 404 is connected to the above-mentioned first acquisition module 402, and is used to acquire target electrochemical impedance spectroscopy data of a second type of target lithium battery under an excitation current of a predetermined frequency; an input module 406 is connected to the above-mentioned second acquisition module 404, and is used to input the target electrochemical impedance spectroscopy data into the target model to obtain the charging state of the target lithium battery.

[0066] It should be noted here that the above-mentioned first acquisition module 402, second acquisition module 404 and input module 406 correspond to steps S102 to S106 in the embodiment, and the instances and application scenarios implemented by multiple modules and corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned embodiment.

[0067] As an optional embodiment, the above-mentioned first electrochemical impedance spectroscopy data set is obtained based on the following method: charging a first type of lithium battery in a fully discharged state; when multiple charging states are reached, the first type of lithium battery is allowed to stand for a predetermined time; and an excitation current of a predetermined frequency is applied to the first type of lithium battery after standing, and the first electrochemical impedance spectroscopy data set is obtained when multiple charging states are reached.

[0068] As an optional embodiment, the above-mentioned second electrochemical impedance spectroscopy data set is obtained based on the following method: charging the second type of lithium battery in a fully discharged state; when multiple charging states are reached, applying an excitation current of a predetermined frequency to the second type of lithium battery respectively, and obtaining the second electrochemical impedance spectroscopy data set when multiple charging states are reached.

[0069] As an optional embodiment, the above-mentioned target model is obtained based on the following training method, including: constructing an initial model; pre-training the initial model based on the first electrochemical impedance spectroscopy data set, determining the weights and bias values ​​of multiple hidden layers of the initial model, and obtaining an intermediate model, wherein the weights and bias values ​​of the multiple hidden layers are used to characterize the first relationship between the electrochemical impedance spectroscopy data and the charging state of the first type of lithium battery; retraining the intermediate model based on the second electrochemical impedance spectroscopy data set and part of the first electrochemical impedance spectroscopy data set, adjusting the weights and bias values ​​of the last hidden layer of the intermediate model, and obtaining a target model, wherein the weights and bias values ​​of the other hidden layers except the last hidden layer in the multiple hidden layers and the adjusted last hidden layer are used to characterize the second relationship between the electrochemical impedance spectroscopy data and the charging state of the second type of lithium battery, and the adjustment of the weights and bias of the last hidden layer is used to characterize the difference between the first relationship and the second relationship.

[0070] As an optional embodiment, the first electrochemical impedance spectroscopy data set includes a normalized first resistance data set and a first reactance data set, and the second electrochemical impedance spectroscopy data set includes a normalized second resistance data set and a second reactance data set.

[0071] As an optional embodiment, the difference between the number of data in the first electrochemical impedance spectroscopy data set and the number of data in the second electrochemical impedance spectroscopy data set is greater than a predetermined threshold.

[0072] According to an embodiment of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned methods for determining the charging status of a lithium battery.

[0073] According to an embodiment of the present invention, an electronic device is provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the program executes any one of the above-mentioned methods for determining the charging state of a lithium battery when running.

[0074] According to an embodiment of the present invention, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, the steps of any one of the above methods are implemented.

[0075] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0076] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0077] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0078] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0079] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0080] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc. Various media that can store program codes.

[0081] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for determining the charging state of a lithium battery, characterized in that: include: Obtaining a target model, wherein the target model is pre-trained based on a first electrochemical impedance spectroscopy data set and retrained based on a second electrochemical impedance spectroscopy data set, wherein the first electrochemical impedance spectroscopy data set is a set of electrochemical impedance spectroscopy data of a first type of lithium battery under an excitation current of a predetermined frequency, and the second electrochemical impedance spectroscopy data set is a set of electrochemical impedance spectroscopy data of a second type of lithium battery under an excitation current of the predetermined frequency; Acquiring target electrochemical impedance spectroscopy data of the second type of target lithium battery under the excitation current of the predetermined frequency; The target electrochemical impedance spectroscopy data is input into the target model to obtain the charging state of the target lithium battery.

2. The method according to claim 1, characterized in that The first electrochemical impedance spectroscopy data set is obtained based on the following method: charging the first type of lithium battery in a fully discharged state; When multiple charging states are reached, allowing the first type of lithium battery to stand for a predetermined time respectively; An excitation current of the predetermined frequency is applied to each of the first type of lithium batteries after being left at rest, and the first electrochemical impedance spectroscopy data set is obtained when the batteries reach multiple charging states.

3. The method according to claim 1, characterized in that The second electrochemical impedance spectroscopy data set is obtained based on the following method: charging the second type of lithium battery in a fully discharged state; When multiple charging states are reached, the excitation current of the predetermined frequency is applied to the second type of lithium battery respectively, and the second electrochemical impedance spectroscopy data set is obtained when the multiple charging states are reached.

4. The method according to claim 1, wherein The target model is obtained based on the following training methods, including: Build an initial model; Pre-training the initial model based on the first electrochemical impedance spectroscopy data set, determining weights and bias values ​​of multiple hidden layers of the initial model, and obtaining an intermediate model, wherein the weights and bias values ​​of the multiple hidden layers are used to characterize a first relationship between the electrochemical impedance spectroscopy data and the state of charge of the first type of lithium battery; Based on the second electrochemical impedance spectroscopy data set and part of the first electrochemical impedance spectroscopy data set, the intermediate model is retrained, and the weights and bias values ​​of the last hidden layer of the intermediate model are adjusted to obtain the target model, wherein the weights and bias values ​​of the other hidden layers in the multiple hidden layers except the last hidden layer and the adjusted last hidden layer are used to characterize the second relationship between the electrochemical impedance spectroscopy data and the charge state of the second type of lithium battery, and the adjustment of the weights and bias of the last hidden layer is used to characterize the difference between the first relationship and the second relationship.

5. The method according to claim 1, wherein The first electrochemical impedance spectroscopy data set includes a normalized first resistance data set and a first reactance data set, and the second electrochemical impedance spectroscopy data set includes a normalized second resistance data set and a second reactance data set.

6. The method according to any one of claims 1 to 5, characterized in that A difference between the number of data in the first electrochemical impedance spectroscopy data set and the number of data in the second electrochemical impedance spectroscopy data set is greater than a predetermined threshold.

7. A device for determining the charging state of a lithium battery, characterized in that: include: a first acquisition module, configured to acquire a target model, wherein the target model is pre-trained based on a first electrochemical impedance spectroscopy data set and re-trained based on a second electrochemical impedance spectroscopy data set, wherein the first electrochemical impedance spectroscopy data set is a set of electrochemical impedance spectroscopy data of a first type of lithium battery under an excitation current of a predetermined frequency, and the second electrochemical impedance spectroscopy data set is a set of electrochemical impedance spectroscopy data of a second type of lithium battery under an excitation current of the predetermined frequency; A second acquisition module is used to obtain target electrochemical impedance spectroscopy data of the second type of target lithium battery under the excitation current of the predetermined frequency; An input module is used to input the target electrochemical impedance spectroscopy data into the target model to obtain the charging state of the target lithium battery.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the lithium battery charging state determination method according to any one of claims 1 to 6.

9. An electronic device, characterized in that: include: a memory storing an executable program; A processor is used to run the program, wherein the program executes the lithium battery charging state determination method according to any one of claims 1 to 6 when running.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images