Battery monitoring method, medium and electronic equipment

Abstract

The invention provides a battery monitoring method, a medium and electronic equipment, and relates to the field of data processing. The battery monitoring method comprises the following steps: obtaining detection parameters of a battery; the detection parameters are obtained by collecting a plurality of charging parameters of the battery in different charging stages; the detection parameters are input into a prediction model to obtain long-time characteristics and short-time characteristics of the detection parameters, the long-time characteristics are used for representing aging information of the battery in a first time period, the short-time characteristics are used for representing aging information of the battery in a second time period, and the long-time characteristics are used for representing aging information of the battery in a third time period; the first time period is far greater than the second time period; obtaining prediction parameters of the battery based on the long-time characteristics and the short-time characteristics; the prediction parameter is used for representing the health state of the battery.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and in particular to a battery monitoring method, medium, and electronic device. Background Art

[0002] During use, batteries will experience performance degradation due to electrochemical side reactions and material aging. Accurately monitoring the health status of batteries is crucial to ensuring safe operation of equipment and extending battery life. However, the battery aging process involves complex changes in timing characteristics, and related battery monitoring methods have difficulty effectively handling these complex characteristics. Summary of the Invention

[0003] In view of this, the present disclosure provides a battery monitoring method, a medium, and an electronic device.

[0004] According to a first aspect of the present disclosure, a battery monitoring method is provided, comprising:

[0005] Obtaining detection parameters of the battery; the detection parameters are obtained by collecting multiple charging parameters of the battery at different charging stages;

[0006] Inputting the detection parameters into a prediction model to obtain long-term and short-term features of the detection parameters, wherein the long-term features are used to characterize aging information of the battery in a first time period, and the short-term features are used to characterize aging information of the battery in a second time period, wherein the first time period is much longer than the second time period;

[0007] A prediction parameter of the battery is obtained based on the long-term characteristics and the short-term characteristics; the prediction parameter is used to characterize the health state of the battery.

[0008] According to an embodiment of the present disclosure, the prediction model includes a parallel structure, which is used to extract the long-term features and the short-term features of the detection parameters, and output a fusion feature obtained by fusing the long-term features and the short-term features;

[0009] The prediction parameter is obtained based on the fusion feature.

[0010] According to an embodiment of the present disclosure, the prediction model includes a serial structure connected to the parallel structure, and the serial structure is used to smooth the fusion features to obtain target features; the target features are used to perform regression prediction to obtain prediction parameters.

[0011] According to an embodiment of the present disclosure, the parallel structure includes a first self-attention layer and a first gated recurrent unit, the first self-attention layer is used to extract the long-term features; the first gated recurrent unit is used to extract the short-term features; the series structure includes a second gated recurrent unit, and the second gated unit is connected in series to process the fusion features.

[0012] According to an embodiment of the present disclosure, obtaining the detection parameters of the battery includes:

[0013] When the charging stage is a full-charge stage, the number of cycles of the battery at the time of full charge is collected. The number of cycles represents the cumulative degree of aging of the battery and is used to provide a timing benchmark for the prediction model.

[0014] According to an embodiment of the present disclosure, obtaining the detection parameters of the battery includes:

[0015] When the charging stage is a constant voltage charging stage, collecting the constant voltage charging time and constant voltage charging capacity of the battery;

[0016] The constant-voltage charging time is used to characterize the change in the polarization degree of the battery. The growth trend of the constant-voltage charging time of the battery with the number of cycles is extracted as the long-term feature through the parallel structure; the change of the constant-voltage charging time between adjacent cycles is extracted as the short-term feature through the parallel structure.

[0017] The constant-voltage charging capacity is used to characterize the degree of active material loss of the battery. The attenuation trend of the constant-voltage charging capacity is extracted as the long-term feature through the parallel structure, and the change of the constant-voltage charging capacity within a single cycle is extracted as the short-term feature through the parallel structure.

[0018] According to an embodiment of the present disclosure, the obtaining of battery detection parameters includes:

[0019] When the charging stage is a full-charge rest stage, collecting the rest start voltage and rest end voltage of the battery;

[0020] When the charging stage is a discharge start stage, collecting the discharge start voltage of the battery;

[0021] The shelf starting voltage is used to characterize the self-discharge characteristics of the battery. The variation trend of the shelf starting voltage with the number of cycles is extracted as the long-term feature through the parallel structure, and the difference of the shelf starting voltage between adjacent cycles is extracted as the short-term feature through the parallel structure.

[0022] The shelf termination voltage is used to characterize the polarization relaxation degree of the battery. The long-term decay trend of the shelf termination voltage is extracted as the long-term feature through the parallel structure, and the change of the shelf termination voltage during the static process is extracted as the short-term feature through the parallel structure.

[0023] The discharge start voltage is used to characterize the change in the internal resistance of the battery. The downward trend of the discharge start voltage with the aging process is extracted as the long-term feature through the parallel structure, and the sudden change of the discharge start voltage at the moment of discharge is extracted as the short-term feature through the parallel structure.

[0024] According to an embodiment of the present disclosure, the method further includes:

[0025] generating a characteristic matrix based on the detection parameters, wherein the characteristic matrix is ​​used to characterize the battery state of the battery at a previous moment and a current moment;

[0026] Position encoding is performed on each sequence in the feature matrix, and the encoded feature matrix is ​​input into a prediction model to obtain the long-term features and short-term features of the detection parameters.

[0027] The second aspect of the present application further provides a computer-readable storage medium having a computer program or instructions stored thereon, which implements the steps of the above method when the computer program or instructions are executed by a processor.

[0028] A third aspect of the present disclosure provides an electronic device, including:

[0029] A collection device, used to collect multiple parameters of the battery at different charging stages to obtain detection parameters;

[0030] A processor is configured to obtain detection parameters of the battery; input the detection parameters into a prediction model to obtain long-term and short-term characteristics of the detection parameters, wherein the long-term characteristics are used to characterize aging information of the battery in a first time period, and the short-term characteristics are used to characterize aging information of the battery in a second time period, and the first time period is much longer than the second time period; and obtain prediction parameters of the battery based on the long-term and short-term characteristics; the prediction parameters are used to characterize the health status of the battery.

[0031] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The above and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0033] Figure 1 A flowchart of a battery monitoring method provided by an embodiment of the present disclosure is schematically shown;

[0034] Figure 2 A schematic diagram of the structure of a prediction model provided by an embodiment of the present disclosure is shown;

[0035] Figure 3 A comparison chart of single model prediction results based on the first self-attention layer is shown for the embodiment of the present disclosure;

[0036] Figure 4 A comparison diagram of prediction results of a single model based on a first gated recurrent unit provided by an embodiment of the present disclosure is schematically shown;

[0037] Figure 5 Schematic diagram of a single model prediction error analysis based on the first self-attention layer provided by an embodiment of the present disclosure

[0038] Figure 6 Schematically shows a single model prediction error analysis diagram based on the first gated recurrent unit provided by an embodiment of the present disclosure;

[0039] Figure 7 A diagram schematically shows a prediction model training running speed performance diagram provided by an embodiment of the present disclosure;

[0040] Figure 8 A diagram schematically shows a comparison between a training set and a prediction result provided by an embodiment of the present disclosure;

[0041] Figure 9 A comparison diagram of prediction results of a test set provided by an embodiment of the present disclosure is schematically shown;

[0042] Figure 10 A test set prediction error analysis diagram provided by an embodiment of the present disclosure is schematically shown;

[0043] Figure 11 A performance evaluation diagram of a coefficient of determination of a prediction model provided by an embodiment of the present disclosure is schematically shown;

[0044] Figure 12 A characteristic evolution diagram of a battery constant voltage charging time as a function of cycle number provided by an embodiment of the present disclosure is schematically shown;

[0045] Figure 13 A diagram schematically shows a capacity decay characteristic diagram of a battery constant voltage charging capacity as a function of cycle number provided by an embodiment of the present disclosure;

[0046] Figure 14 A voltage variation characteristic curve diagram of a battery provided by an embodiment of the present disclosure during a discharge process after full charge is schematically shown;

[0047] Figure 15 A diagram schematically illustrates an importance analysis of battery health prediction features based on a GRU model provided in an embodiment of the present disclosure;

[0048] Figure 16 The block diagram of an electronic device provided by an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0049] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0050] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0051] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0052] When expressions such as "at least one of A, B, and C, etc." are used, they should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0053] In the embodiments of this disclosure, the collection, updating, analysis, processing, use, transmission, provision, disclosure, and storage of all data involved (including, but not limited to, user personal information) comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures have been taken to prevent unauthorized access to user personal information data and to safeguard the security of user personal information, network security, and national security.

[0054] The present disclosure provides a battery monitoring method, medium, and electronic device. The battery monitoring method includes:

[0055] Obtaining battery detection parameters; the detection parameters are obtained by collecting multiple charging parameters of the battery at different charging stages;

[0056] Inputting the detection parameters into a prediction model to obtain long-term and short-term features of the detection parameters, wherein the long-term features are used to characterize aging information of the battery in a first time period, and the short-term features are used to characterize aging information of the battery in a second time period, where the first time period is much longer than the second time period;

[0057] The battery's predicted parameters are obtained based on the long-term and short-term features; the predicted parameters are used to characterize the battery's health status.

[0058] By adopting the embodiment of the present disclosure, detection parameters are obtained by collecting multiple charging parameters of the battery in different charging stages, and characteristic information of the battery under multiple operating states can be obtained, providing a rich data basis for subsequent feature extraction. After the detection parameters are input into the prediction model, long-term features and short-term features can be obtained at the same time, wherein the long-term features characterize the aging information of the battery in the first time period, and the short-term features characterize the aging information of the battery in the second time period. Since the first time period is much larger than the second time period, the prediction model can capture the long-term trend changes and short-term fluctuation changes in the battery aging process, respectively, thereby comprehensively characterizing the complex time series characteristics of battery aging. The prediction parameters obtained based on long-term features and short-term features can comprehensively reflect the health status information of the battery at different time scales. Compared with the problem that it is difficult to effectively handle complex time series features in related technologies, the embodiment of the present disclosure can more accurately monitor the battery status, thereby effectively predicting the health status of the battery.

[0059] The electronic devices in the embodiments of the present disclosure refer to intelligent devices with battery monitoring, management, and analysis capabilities. Such electronic devices include consumer electronics such as smartphones, tablets, and laptops; new energy vehicles such as electric vehicles, hybrid vehicles, and electric bicycles; and energy storage devices equipped with battery management systems, uninterruptible power supplies, mobile power supplies, and other energy storage devices. In addition, battery health monitoring modules for smart terminals such as drones, robots, and wearable devices; professional battery management components such as backup power management systems and base station backup power status monitoring units in industrial equipment; vertical battery application units such as portable monitor power modules in medical equipment, independent power supply units for security systems, and battery life managers for communication equipment also fall within the scope of the electronic devices described in the embodiments of the present disclosure. Furthermore, special scenario battery devices such as smart sensor nodes in IoT environments, battery status monitoring centers for smart home devices, and coordinated management devices for distributed energy storage systems, as well as distributed battery management systems such as cloud-based battery data analysis servers, edge computing battery management units, and monitoring nodes for layered architecture energy storage systems, are all within the technical adaptation scope of the embodiments of the present disclosure.

[0060] It should be noted that the embodiments of the present disclosure do not limit the specific types of electronic devices. The above examples are only illustrative descriptions. The technical solutions of the present disclosure can be applied to any electronic device with battery monitoring, management and analysis capabilities.

[0061] The following describes a battery monitoring method provided in an embodiment of the present disclosure.

[0062] Figure 1 The figure schematically shows a flow chart of a battery monitoring method provided by an embodiment of the present disclosure.

[0063] like Figure 1 As shown, the battery monitoring method may further include operations S101 to S103.

[0064] Operation S101: obtaining detection parameters of a battery; the detection parameters are obtained by collecting multiple charging parameters of the battery at different charging stages;

[0065] Operation S102: Inputting the detection parameters into a prediction model to obtain long-term and short-term features of the detection parameters, wherein the long-term features are used to characterize aging information of the battery in a first time period, and the short-term features are used to characterize aging information of the battery in a second time period, where the first time period is much longer than the second time period.

[0066] In operation S103 , a prediction parameter of the battery is obtained based on the long-term characteristics and the short-term characteristics; the prediction parameter is used to characterize the health state of the battery.

[0067] In operation S101, the detection parameters refer to data indicators that can reflect the battery's health status and aging level after feature extraction and data processing. In the disclosed embodiments, these can be understood as characteristic variables with battery aging sensitivity and predictive value extracted from the raw monitoring data of the battery charging process. These are used as inputs to the prediction model to accurately predict and evaluate the battery's health.

[0068] Exemplarily, the detection parameters include, but are not limited to: time parameters, capacity parameters, voltage parameters, current parameters, resistance parameters, efficiency parameters, temperature parameters, cycle parameters, power parameters, etc.

[0069] Furthermore, a charging phase refers to a time period with different operating characteristics during a battery's complete charging cycle, naturally divided according to the charging strategy and the battery's electrochemical characteristics. In the disclosed embodiments, this can be understood as a phased time window during the battery charging process that exhibits different voltage and current variation patterns, used to collect differentiated charging data under specific battery operating conditions.

[0070] Exemplarily, the charging stages include but are not limited to a charging preparation stage, a constant current charging stage, a constant voltage charging stage, a charging completion stage, a full charge rest stage, a discharge preparation stage, a constant current discharge stage, a constant voltage discharge stage, a discharge completion stage, a post-discharge rest stage, and the like.

[0071] Similarly, charging parameters refer to physical quantities directly measured by sensors or calculated based on measurement data during the battery charging process. In the disclosed embodiments, these can be understood as observable data variables that describe the electrochemical behavior and operating state of the battery during charging, and are used to quantitatively characterize the performance and state characteristics of the battery at different charging stages.

[0072] It is worth noting that the selection of charging parameters in the embodiments of the present disclosure is based on the common mechanisms of battery electrochemical behavior, rather than the individual characteristics of specific materials, thereby achieving technical compatibility across material systems. Specifically, the selected charging parameters reflect the basic physical and chemical phenomena of the battery during the charging and discharging process, such as polarization behavior, ion transfer impedance, and active material utilization. These phenomena are common in different negative electrode materials (such as graphite negative electrodes, silicon negative electrodes, etc.) and positive electrode material systems, and differ only in the numerical range and change trend, without changing their validity as health indicator parameters.

[0073] Furthermore, the charging parameters used in the embodiments of the present disclosure have the technical advantage of non-invasive measurement. All parameters can be obtained through standardized sensors outside the battery, including but not limited to conventional monitoring equipment such as voltage sensors, current sensors, and temperature sensors, without the need for special equipment or laboratory analysis methods. This measurement method avoids destructive testing operations such as battery disassembly, electrolyte sampling, and electrode material separation, maintaining the integrity of the battery structure and the continuity of its use function.

[0074] Furthermore, parameter acquisition methods based on conventional sensors offer engineering advantages. Voltage and current sensors, standard components of battery management systems, can be deployed on a large scale in applications such as consumer electronics, electric vehicles, and energy storage systems without incurring additional hardware costs. Furthermore, these sensors offer fast response times, high measurement accuracy, and excellent long-term stability, enabling real-time monitoring of charging parameters.

[0075] In one feasible implementation, the electronic device obtains detection parameters via a built-in battery management chip. The battery management chip integrates voltage and current sensors and converts analog signals into digital signals using an ADC. The electronic device's main processor communicates with the battery management chip via a bus, reading voltage and current data in real time. It then extracts the corresponding charging parameters based on a charging stage recognition algorithm and, after preprocessing, generates the detection parameters.

[0076] In another feasible embodiment, the electronic device acquires detection parameters via an external sensor module. The external sensor module includes a clamp-type current sensor and a contact-type voltage sensor, which connects to the electronic device via a USB or Bluetooth interface. The electronic device runs parameter extraction software to automatically identify charging phase switching points based on the sensor data and collect charging parameters at critical moments.

[0077] In operation S102, the prediction model refers to a processing model built based on a machine learning algorithm, which performs feature extraction and pattern recognition on input detection parameters to achieve a quantitative assessment of battery aging and health status. In the disclosed embodiment, it can be understood as a neural network model using a deep learning architecture.

[0078] Exemplarily, the prediction model includes but is not limited to: a recurrent neural network model, a convolutional neural network model, an attention mechanism model, an ensemble learning model, a support vector machine model, etc.

[0079] Furthermore, long-term characteristics refer to characteristic variables that reflect the cumulative changes and progressive weakening of a battery over a long period of time. Their changing patterns reflect the inherent aging patterns of battery materials and the effects of long-term use. In the disclosed embodiments, these can be understood as data features extracted from the time series of detection parameters that characterize the long-term aging trends of the battery. These can capture long-term changes in the battery's performance degradation patterns, capacity decline trends, impedance growth patterns, and other long-term changes throughout its entire lifespan, and are used to establish a correlation between the battery's current health status and its historical usage.

[0080] Similarly, short-term features refer to characteristic variables that reflect the battery's transient response and dynamic changes within a relatively short time window. Their variation patterns reflect the battery's immediate performance and local state fluctuations under specific operating conditions. In the disclosed embodiments, these can be understood as data features that characterize short-term battery state changes, extracted from a local time period of the detection parameters. These can capture short-term characteristics such as voltage mutations, current fluctuations, and power changes during a single charge and discharge cycle, and are used to identify short-term anomalies and transient changes in the battery's state.

[0081] At the same time, long-term and short-term features are complementary and synergistic in battery health prediction. Long-term features are used to characterize battery aging information in the first time period, while short-term features are used to characterize battery aging information in the second time period. Long-term features provide the macroscopic background and overall trend information of battery aging, while short-term features provide microscopic details and local changes in battery status. The combination of the two forms a comprehensive, multi-scale representation of battery health status.

[0082] It's important to note that the first time period is significantly longer than the second time period, differing by orders of magnitude. Specifically, the first time period is typically measured in terms of the number of complete charge and discharge cycles or the length of time a battery has been in use, reflecting the battery's long-term aging process. The second time period, typically measured in terms of the duration of a battery's specific charging phase or the time interval between adjacent measurement points, reflects the battery's instantaneous state changes.

[0083] For example, the first time period could be 100-1000 complete charge-discharge cycles experienced by the battery, corresponding to a battery life span of several months to several years. The second time period could be the duration of a specific phase in a single charge cycle, such as a few minutes to tens of minutes during constant-voltage charging, or the time interval between two consecutive parameter acquisitions. This differentiated timescale design allows the prediction model to capture both long-term trends and short-term fluctuations in battery aging, enabling a comprehensive assessment of the battery's health status.

[0084] In operation S103 , a prediction parameter of the battery is obtained based on the long-term characteristics and the short-term characteristics; the prediction parameter is used to characterize the health state of the battery.

[0085] In operation S103, the prediction parameter refers to a quantitative indicator output after comprehensive analysis and calculation of long-term and short-term features using a prediction model. It is used to directly reflect the current health level and degree of aging of the battery. In the disclosed embodiments, it can be understood as a numerical result representing the battery health status obtained through feature fusion and regression calculation. It can convert complex battery aging information into a health assessment indicator that is easy to understand and apply, thereby providing accurate status information to the battery management system and intuitive health status feedback to the user.

[0086] Exemplarily, the prediction parameters include, but are not limited to: battery health percentage, remaining available capacity, expected remaining life, aging index, performance degradation rate, etc.

[0087] In a feasible implementation, prediction parameters can be obtained by weighted fusion of features. Specifically, weight coefficients are assigned to long-term features and short-term features respectively, and the weight distribution ratio is dynamically adjusted according to the current use stage and aging degree of the battery. In the early stage of battery use, the weight of short-term features is higher, and the focus is on the immediate performance of the battery; in the later stage of battery use, the weight of long-term features is higher, and the focus is on the cumulative aging trend of the battery. The long-term features and short-term features are merged into fused features by weighted summation, and then the fused features are mapped to prediction parameters through the regression layer.

[0088] By adopting the embodiment of the present disclosure, detection parameters are obtained by collecting multiple charging parameters of the battery in different charging stages, and characteristic information of the battery under multiple operating states can be obtained, providing a rich data basis for subsequent feature extraction. After the detection parameters are input into the prediction model, long-term features and short-term features can be obtained at the same time, wherein the long-term features characterize the aging information of the battery in the first time period, and the short-term features characterize the aging information of the battery in the second time period. Since the first time period is much larger than the second time period, the prediction model can capture the long-term trend changes and short-term fluctuation changes in the battery aging process, respectively, thereby comprehensively characterizing the complex time series characteristics of battery aging. The prediction parameters obtained based on long-term features and short-term features can comprehensively reflect the health status information of the battery at different time scales. Compared with the problem that it is difficult to effectively handle complex time series features in related technologies, the embodiment of the present disclosure can more accurately monitor the battery status, thereby effectively predicting the health status of the battery.

[0089] The following is an introduction to the prediction model provided by the embodiments of the present disclosure.

[0090] In practical applications, the prediction models of related technologies can usually only effectively process characteristic information of a single time scale. They either focus on long-term trends and ignore the warning value of short-term changes, or focus on short-term fluctuations and lack a grasp of the overall aging process of the battery, resulting in a difficult balance between long-term stability and short-term sensitivity in the prediction results.

[0091] In response to the above technical problems, based on the above embodiments, as an optional embodiment, the prediction model includes a parallel structure, which is used to extract the long-term features and short-term features of the detection parameters, and output the fusion of the long-term features and the short-term features to obtain the fusion features; the prediction parameters are obtained based on the fusion features.

[0092] Specifically, the disclosed embodiments employ a prediction model with a parallel structure to simultaneously extract both long-term and short-term features of detection parameters. In this technical solution, the parallel structure of the prediction model is designed as two parallel feature extraction channels. Both channels receive the same detection parameter inputs but utilize different feature extraction mechanisms to process information at different time scales.

[0093] Specifically, the first channel of the parallel structure is used to extract long-term features. This channel adopts an algorithm architecture that can model long-distance dependencies. Through the global attention mechanism, the complete time series of detection parameters is analyzed to identify the long-term change characteristics of the battery such as the cumulative aging pattern, capacity attenuation trend and impedance growth law throughout the entire usage cycle.

[0094] The second channel of the parallel structure is used to extract short-term features. This channel adopts an algorithm architecture that can capture local timing changes. Through the sequence modeling mechanism, it analyzes the changes of detection parameters between adjacent time points and identifies the short-term dynamic characteristics of the battery such as voltage mutation, capacity fluctuation and time change in specific charging and discharging stages.

[0095] After completing feature extraction, the two parallel channels fuse the output long-term and short-term features to obtain fused features. This fusion process can be achieved through feature concatenation, where the long-term and short-term feature vectors are concatenated along the feature dimension to form a comprehensive feature representation containing information from multiple time scales. The fused features simultaneously retain both long-term trend information and short-term variation information about battery aging, enabling the prediction model to understand current short-term state changes based on the battery's historical aging context and adjust its assessment of the battery's long-term aging trend based on current short-term state changes. After obtaining the fused features, the prediction model maps the fused features into prediction parameters through regression calculations.

[0096] In the disclosed embodiments, the parallel design allows two feature extraction channels to operate in parallel, ensuring that both long-term and short-term features can extract the most complete information from the original detection parameters. Furthermore, the parallel structure enables the prediction model to simultaneously address both macroscopic trends and microscopic changes in battery aging, forming a comprehensive, multi-scale representation of the battery's health status, improving prediction accuracy and stability.

[0097] While the parallel structure can effectively extract long-term and short-term features and obtain fused features, fused features often contain conflicts and noise between information at different time scales. Especially when the battery operates under special operating conditions, instantaneous fluctuations in short-term features may conflict with the stable trends in long-term features, resulting in numerical jumps and discontinuous changes in the fused features. Furthermore, regression prediction directly based on fused features is susceptible to feature dimension mismatch and uneven data distribution, making the prediction results unstable over time series.

[0098] In response to the above technical problems, based on the above embodiments, as an optional embodiment, the prediction model includes a serial structure connected to a parallel structure, the serial structure is used to smooth the fused features to obtain target features, and the target features are used to perform regression prediction to obtain prediction parameters.

[0099] Specifically, the disclosed embodiment adds a serial structure to the parallel structure, optimizing the quality and stability of fused features through a serialized feature processing flow. The serial structure receives the fused features output by the parallel structure as input and uses a temporal smoothing algorithm to coordinate the multi-timescale information in the fused features.

[0100] During the smoothing process of the series structure, the fusion features are first analyzed in time series to identify the trend components, periodic components and random components. The high-frequency noise and abnormal mutations are removed through the filtering mechanism, and the effective information that is indicative of the battery health status is retained.

[0101] For example, the smoothing algorithm employs an adaptive weight allocation strategy, dynamically adjusting the smoothing strength based on the credibility of the long-term and short-term features in the fused features. Weaker smoothing is used during stable phases dominated by long-term features to maintain trend sensitivity, while stronger smoothing is used during anomalous phases with large short-term feature fluctuations to suppress noise interference. Through these differentiated smoothing strategies, the cascade structure improves the temporal continuity of the data and the stability of predictions while maintaining the integrity of feature information.

[0102] After smoothing, the cascade structure outputs target features, which have better data quality and greater predictive applicability than fused features. Target features eliminate conflicts and contradictions between information at different time scales, forming a unified feature representation space. This allows subsequent regression predictions to be modeled based on a consistent data distribution. Furthermore, target features preserve the core content of battery aging information, encompassing both the cumulative aging trends in long-term features and the key state changes in short-term features.

[0103] The disclosed embodiments smooth the fused features through a series structure, effectively resolving information conflicts and data instability issues during multi-timescale feature fusion. The resulting target features exhibit improved temporal continuity. Furthermore, the combined design of series and parallel structures ensures comprehensive feature extraction while improving the stability of feature quality. This allows regression prediction based on the target features to obtain more accurate and reliable prediction parameters, thereby enhancing the overall performance of battery health monitoring.

[0104] While the combination of parallel and series architectures enables the extraction and optimization of multi-timescale features, different algorithm modules have different modeling capabilities and applicable scenarios when processing battery aging information. Traditional prediction models often use a single type of neural network architecture, which makes it difficult to simultaneously meet the dual requirements of modeling long-term dependencies and capturing short-term sequence changes.

[0105] In response to the above technical problems, on the basis of the above embodiments, as an optional embodiment, the parallel structure includes a first self-attention layer and a first gated recurrent unit, the first self-attention layer is used to extract long-term features, and the first gated recurrent unit is used to extract short-term features. The series structure includes a second gated recurrent unit, and the second gated unit is connected in series to process fusion features.

[0106] Figure 2 The figure schematically shows a structural diagram of a prediction model provided by an embodiment of the present disclosure.

[0107] like Figure 2 As shown, the detection parameters 201 are simultaneously input into the first self-attention layer 203 and the first gated recurrent unit 204 in the parallel structure. The first self-attention layer 203 is used to extract long-term features, which uses a multi-head attention mechanism to perform a global analysis on the complete time series of the detection parameters 201.

[0108] By calculating the correlation between each time position in the sequence and all other time positions, the self-attention mechanism can directly establish dependencies across long time spans, eliminating the need for gradual information transfer as in traditional recurrent networks. When processing the long-term characteristics of batteries, the first self-attention layer 203 can identify correlation patterns between batteries at different cycle times, capturing long-term variations such as the decay trend of battery capacity with cycle number and the growth pattern of internal resistance with aging. The parallel computing nature of the self-attention mechanism allows long-term feature extraction to be unrestricted by sequence length, enabling effective processing of data from the long-term battery usage history.

[0109] At the same time, the first gated cycle unit 204 is used to extract short-term features, which uses a gating mechanism to dynamically model the local time series of the detection parameters 201. The gated cycle unit can selectively retain and forget information at different time steps through the control mechanism of resetting the gate and updating the gate, and is particularly suitable for processing short-term changes and local fluctuations in the sequence. When processing the short-term characteristics of the battery, the first gated cycle unit 204 can sensitively capture the state changes of the battery between adjacent charge and discharge cycles, and identify short-term dynamic features such as fluctuations in constant voltage charging time and jumps in shelf voltage. The adaptive selection capability of the gating mechanism enables short-term feature extraction to adjust the degree of dependence on historical information according to the current state of the battery, and highlight key state change information while maintaining time series continuity.

[0110] After the first self-attention layer 203 and the first gated recurrent unit 204 extract long-term and short-term features, respectively, they are fused together to form a fused feature through concatenation 205. Concatenation 205 concatenates the long-term and short-term feature vectors along the feature dimension, forming a comprehensive feature representation that encompasses information at multiple timescales. The fused feature encompasses both global trends and local variations in battery aging.

[0111] The second gated recurrent unit 206 in the cascade structure receives the fused features as input and is used to address multi-timescale information conflicts and data instability within the fused features. The second gated recurrent unit 206 can utilize the same gating mechanism as the first gated recurrent unit 204, but its processing objective differs from short-term feature extraction. Instead, it performs temporal smoothing and information coordination on the fused multi-scale features.

[0112] The second gated recurrent unit 206 dynamically adjusts the weight distribution of information of different time scales in the fusion features through a gated selection mechanism. For time points where there is a conflict between long-term features and short-term features, information with higher credibility is retained first, while suppressing the influence of noise and abnormal fluctuations.

[0113] After processing by the second gated recurrent unit 206, the features enter the rectified linear unit 207 for nonlinear activation, enhancing the model's feature representation capabilities. Regularization is then performed through dropout 208 to prevent overfitting and improve generalization performance. The fully connected layer 209 maps the processed features to the prediction output space, ultimately yielding predicted parameters representing the battery's health status through regression output 210.

[0114] In an optional embodiment, the first self-attention layer 203 may include one or more self-attention layers, each of which uses a multi-head attention mechanism for parallel feature extraction. When the first self-attention layer 203 includes multiple self-attention layers, self-attention layers at different levels can extract long-term features of different levels of abstraction. Shallow self-attention layers focus on extracting direct temporal correlation features of battery parameters, while deep self-attention layers are responsible for extracting more abstract long-term aging pattern features. The multi-layer self-attention structure can more comprehensively capture the complex aging laws of batteries over a long period of time through layer-by-layer feature abstraction and information transfer, thereby improving the accuracy and completeness of long-term feature extraction.

[0115] In another optional embodiment, the series structure may further include a second self-attention layer, which is connected in series with the second gated recurrent unit 206 to perform global correlation analysis on the fused features before temporal smoothing. The second self-attention layer receives the fused features from the splicing 205 as input, analyzes the global correlation between the long-term features and the short-term features in the fused features through the self-attention mechanism, and identifies the consistency and contradiction between information at different time scales. The global modeling capability of the second self-attention layer can evaluate the information quality of the fused features from a holistic perspective, and provide correlation guidance information for the subsequent temporal smoothing of the second gated recurrent unit 206. Through the preprocessing of the second self-attention layer, the second gated recurrent unit 206 can perform more accurate information selection and weight adjustment based on the results of the global correlation analysis, thereby improving the effect of the fused feature processing and the quality of the target features. The second self-attention layer can also include one or more self-attention layers. The multi-layer structure can realize hierarchical analysis and gradual optimization of the fused features, and ensure the coordination and unification of multi-time scale information in the fused features through global correlation modeling at different levels of abstraction.

[0116] Figure 3 A comparison diagram of single model prediction results based on the first self-attention layer is shown for the embodiment of the present disclosure.

[0117] like Figure 3As shown in the figure, a single prediction model constructed using the first self-attention layer is used to compare the health prediction results of the B02 battery and the Transformer model. As can be seen from the figure, within the test range of 0-500 battery cycles, the single model prediction curve based on the first self-attention layer has a good degree of consistency with the actual battery health curve, with a coefficient of determination R² of 0.95491 and a root mean square error (RMSE) of 0.41025, indicating that the first self-attention layer has good modeling capabilities in extracting long-term battery characteristics.

[0118] Figure 4 A comparison diagram of single model prediction results based on the first gated recurrent unit provided in an embodiment of the present disclosure is schematically shown.

[0119] like Figure 4 As shown in the figure, a single prediction model constructed using the first gated recurrent unit was used to compare the health prediction results of the B02 battery and the GRU model. The test results show that the single model based on the first gated recurrent unit also demonstrated good accuracy in battery health prediction, with a coefficient of determination (R²) of 0.95555 and a root mean square error (RMSE) of 0.40735, indicating that the first gated recurrent unit has effective sequence modeling capabilities in capturing short-term changes in battery characteristics.

[0120] Figure 5 A single model prediction error analysis diagram based on the first self-attention layer is shown schematically for an embodiment of the present disclosure.

[0121] like Figure 5 As shown in the figure, the prediction error of the single model based on the first self-attention layer is relatively stable within the battery cycle range of 0-400 cycles, with the error fluctuation range mainly concentrated within ±0.5%. However, after the battery cycle exceeds 400 cycles, the prediction error shows a significant increasing trend, with the cumulative deviation gradually increasing and the maximum error approaching -1.2%. This phenomenon shows that while the use of the first self-attention layer alone can effectively model the long-term aging trend of batteries, it has limitations in dealing with the complex nonlinear changes in the late stage of battery aging, and it is difficult to accurately capture short-term fluctuations and abnormal changes in the battery state.

[0122] Figure 6 A single model prediction error analysis diagram based on the first gated recurrent unit provided in an embodiment of the present disclosure is schematically shown.

[0123] like Figure 6As shown in the figure, the prediction error of the single model based on the first gated recurrent unit during the entire test cycle shows a similar variation pattern to that of the first self-attention layer. In the battery cycle range of 0-350 cycles, the prediction error remains in a relatively small range, and the error fluctuation is mainly within ±0.4%. However, when the battery enters the deep aging stage, the prediction error also increases significantly. In the cycle range of 400-500 cycles, the error accumulation effect is obvious, and the maximum error reaches about -1.1%. This shows that although the use of the first gated recurrent unit alone has advantages in short-term feature extraction, it lacks a comprehensive understanding of the long-term aging background of the battery, and is prone to prediction deviation accumulation problems in the later stages of battery aging.

[0124] From the above results, we can see that the accuracy of the single model in predicting battery health is <1.5%, which is a certain improvement compared to the physical method of capacity integration used in related technologies (<3%). However, the error increases significantly after 400 cycles, and the cumulative deviation increases.

[0125] Figure 7 A performance diagram of the prediction model training running speed provided by an embodiment of the present disclosure is schematically shown.

[0126] like Figure 7 As shown, the prediction model of the embodiment of the present disclosure was trained and tested in a single-core CPU environment, demonstrating the model's computational efficiency and resource consumption. The test results show that the prediction model architecture that combines parallel and series structures can achieve an efficient training process while ensuring prediction accuracy. This model is trained using a single-core CPU with a training speed of approximately 10 minutes. Compared to traditional deep learning models that require hours or even days of training time, the prediction model of the embodiment of the present disclosure has significant computational efficiency advantages. This fast training capability enables the prediction model to adapt to the needs of online learning and model updates in actual application scenarios.

[0127] Figure 8 A comparison diagram of a training set and prediction results provided by an embodiment of the present disclosure is schematically shown.

[0128] like Figure 8 As shown in Figure 2, the prediction results of the prediction model on the training set are compared with the actual battery health data. Figure 8 As can be seen in the figure, the output curve of the prediction model closely matches the actual battery health curve in the training set, indicating that the model successfully learned the basic laws and characteristic patterns of battery aging during training. The good fit of the training set verifies the feature extraction capabilities of the first self-attention layer and the first gated recurrent unit in the parallel structure, as well as the feature fusion and smoothing effects of the second gated recurrent unit in the series structure.

[0129] Figure 9A comparison diagram of test set prediction results provided by an embodiment of the present disclosure is schematically shown.

[0130] like Figure 9 As shown in the figure, the prediction results of the prediction model on the test set B02 battery data are compared with the actual health curve for verification. The test results show that the prediction model can still maintain high-precision prediction performance on test data that has not participated in training, and the deviation between the predicted curve and the true curve is extremely small, proving that the model has good generalization ability and practical application value. Especially in the deep aging stage of 400-500 cycles of the battery cycle, the prediction model can still accurately track the changing trend of the battery health, effectively solving the problem of cumulative deviation of the single model in the late stage of battery aging. This result verifies the effectiveness of the combination of parallel and series structures in processing complex timing characteristics.

[0131] Figure 10 A test set prediction error analysis diagram provided by an embodiment of the present disclosure is schematically shown.

[0132] like Figure 10 As shown in the figure, the prediction error distribution of the prediction model on the test set is analyzed in detail. The error curve shows that the prediction error remains within a small fluctuation range throughout the battery cycle range of 0-500 cycles, without the sharp increase in error observed in the later stages of the single model. The error distribution is relatively uniform, with the prediction error at most time points controlled within ±0.3% and the maximum error not exceeding ±0.5%, indicating that the prediction model has excellent prediction stability and reliability. This level of error control far exceeds the performance of single models in related technologies, fully demonstrating the advantages of the multi-timescale feature fusion technology solution.

[0133] Figure 11 A performance evaluation diagram of the determination coefficient of a prediction model provided by an embodiment of the present disclosure is schematically shown.

[0134] like Figure 11 As shown in the figure, the coefficient of determination R² of the prediction model reaches 0.97968, indicating that the model can explain 97.97% of the variance in the test data, with extremely high fitting accuracy and prediction accuracy. The excellent performance of the coefficient of determination close to 1.0 shows that the prediction model successfully captures the main laws and characteristic patterns of battery health changes, and there is a strong correlation between the model prediction value and the actual value. This performance indicator is significantly better than Figure 3-Figure 6 The R² value of the single model (about 0.955) proves the effectiveness of the technical solution combining parallel and series structures in improving prediction accuracy.

[0135] The prediction model demonstrated excellent performance across multiple evaluation metrics, including mean absolute error (MAE) of 0.23637%, mean bias error (MBE) of 0.17979%, mean square error (MSE) of 0.079051%, root mean square error (RMSE) of 0.28116%, coefficient of determination (R²) of 0.97968, residual prediction error (RPD) of 9.1239, and mean absolute percentage error (MAPE) of 0.0024761. These quantitative metrics fully validate the high accuracy of the prediction model.

[0136] The Transformer-GRU model achieved an accuracy of less than 0.3%, significantly exceeding existing fuel gauge methods that calculate battery health based on physical and chemical properties (with an error of less than 3%). The computation time using a CPU was less than 10 minutes. This experimental setup fully demonstrates the model's ability to estimate battery health with high accuracy, high stability, and low computational effort.

[0137] By adopting the embodiment of the present disclosure, the extraction of long-term and short-term battery features is achieved through the parallel configuration of the first self-attention layer and the first gated recurrent unit. The global modeling capability of the first self-attention layer ensures accurate capture of the long-term aging trend of the battery, and the sequence modeling capability of the first gated recurrent unit ensures sensitive identification of short-term state changes of the battery. The series design of the second gated recurrent unit achieves effective optimization of the fusion feature. Through the information selection and temporal smoothing function of the gating mechanism, the information conflict in the multi-time scale feature fusion process is eliminated, and the reliability of the prediction results is improved.

[0138] In practical application scenarios, prediction models face the problem of missing temporal information when processing detection parameters. Battery aging is a time-dependent physical and chemical process, and detection parameters collected at different times have clear temporal dependencies. For example, the current battery state is often affected by the state at previous times, while future health changes are based on the cumulative effect of the current and historical states. When prediction models lack an understanding of the temporal relationships between detection parameters, they tend to treat data from different time points as equally important feature information, resulting in the model's inability to accurately model the temporal evolution of battery aging.

[0139] In response to the above technical problem, based on the above embodiment, as an optional embodiment, the above battery monitoring method may further include the following operations:

[0140] Operation S201: generating a characteristic matrix based on detection parameters, where the characteristic matrix is ​​used to characterize the battery state at a previous moment and a current moment;

[0141] In operation S202 , position encoding is performed on each sequence in the feature matrix, and the encoded feature matrix is ​​input into a prediction model to obtain long-term features and short-term features of the detection parameters.

[0142] In operation S201, a feature matrix is ​​used to characterize the battery status at the previous moment and the current moment. The feature matrix can be constructed using a sliding window technique to reorganize the detection parameters in the form of a time series into a feature representation in the form of a matrix.

[0143] Specifically, after the system obtains the battery's detection parameters at multiple consecutive moments, it selects two adjacent moments as the basis for constructing a feature matrix. The detection parameters for the previous moment (moment k-1) and the current moment (moment k) are extracted separately, forming a data structure containing information in two time dimensions. Because the detection parameters at each moment contain six eigenvalues, the dimension of the feature matrix is ​​12×1, with the first six elements corresponding to the battery status information at the previous moment and the last six elements corresponding to the battery status information at the current moment. The time series information contained in the feature matrix helps the prediction model understand the changing trends and evolution patterns of the battery status, enabling the model to reason based on historical status information when predicting the battery health at the next moment (moment k+1).

[0144] In operation S202 , when calculating the correlation between different positions in the sequence, the self-attention mechanism essentially distributes attention weights based on content similarity, without considering the absolute position or relative position relationship of the elements in the sequence.

[0145] For battery health prediction, the positional information of the detection parameters has clear physical meaning. Different positions correspond to different time points, and the temporal order directly affects the causal relationship of the battery aging process. Positional encoding explicitly embeds implicit temporal information into the feature representation by assigning a unique encoding vector to each sequence position in the feature matrix.

[0146] In an optional embodiment, the disclosed embodiment adopts a learnable sine-cosine position encoding method, using trigonometric functions of different frequencies to generate distinguishable encoding vectors for different positions. The encoding method based on trigonometric functions can generate smoothly changing encoding vectors with periodic characteristics for different positions, so that the encoding vectors of adjacent positions are similar while ensuring that the encoding vectors of different positions are distinguishable. The position encoding vector is added to the original feature matrix to incorporate the position information into the feature representation, forming a coding feature matrix containing timing constraints.

[0147] When the encoded feature matrix is ​​fed into the prediction model, the self-attention layer simultaneously utilizes both feature content information and positional temporal information to calculate associations. When the self-attention mechanism calculates attention weights between the query vector, key vector, and value vector, the fusion of the position-encoded feature vector enables the model to distinguish detection parameters at different temporal locations and accurately identify the temporal pattern of battery state changes.

[0148] When extracting long-term features, the prediction model accurately understands the cumulative aging trend of the battery over long-term use, avoiding incorrect associations between aging status information from different periods. Simultaneously, when extracting short-term features, it accurately captures the details of the battery's state changes between adjacent moments, ensuring the temporal consistency of short-term fluctuations. The gated recurrent unit also benefits from the timing constraints provided by positional encoding when processing the encoded feature matrix. The gating mechanism selects information and updates status based on clear temporal relationships, improving the accuracy of short-term feature extraction.

[0149] Using the disclosed embodiments, discrete detection parameters are converted into structured temporal feature representations through feature matrix generation technology. This provides the prediction model with input data containing clear temporal relationships, enabling the model to perform comprehensive analysis and prediction based on the battery's historical and current state information. The application of positional encoding technology further enhances the prediction model's temporal modeling capabilities, enabling both the self-attention mechanism and the gated recurrent unit to perform feature extraction and information processing under clear temporal constraints.

[0150] The following introduces the relevant detection parameters in the embodiments of the present disclosure.

[0151] Based on the above embodiment, as an optional embodiment, when the charging stage is the full-charge stage, the number of cycles of the battery at the time of full charge is collected. The number of cycles represents the cumulative degree of battery aging and is used to provide a timing benchmark for the prediction model.

[0152] The cycle count refers to the cumulative count of the number of complete charge and discharge cycles a battery has undergone since its initial state, and is used to quantitatively characterize the cumulative degree of battery aging over its entire lifespan. In the disclosed embodiments, this can be understood as a time-series indicator parameter that reflects the battery's historical usage intensity and aging process. By counting each completed charge and discharge cycle, the battery provides quantitative benchmark information about its current aging state relative to its initial state, enabling feature extraction and analysis of other detection parameters within a unified time-series dimension.

[0153] Specifically, during the battery charging and discharging process, state stability and data reliability vary across different charging stages. The fully charged stage represents the stable state after the battery reaches the set charging cutoff conditions. At this point, the battery's voltage, temperature, and internal electrochemical equilibrium tend to stabilize, providing the optimal decision point for accurate cycle counting.

[0154] In one feasible implementation, the electronic device can automatically collect the number of cycles during the full charge phase through a built-in battery management system. Specifically, the battery management system continuously monitors parameters such as the battery terminal voltage, charging current, and charging time during each charging process. When it detects that the battery voltage reaches the set charging cutoff voltage and the charging current drops to the set cutoff current threshold, the system determines that the battery has reached a full charge.

[0155] After successfully determining a full charge state, the battery management system automatically triggers the cycle counter to accumulate the current cycle count by 1 and stores the updated cycle value in non-volatile memory to ensure data continuity after the device is powered off or restarted. The system also records the timestamp of the cycle count update.

[0156] By using the disclosed embodiments, the battery cycle count at the full charge stage is collected as a timing reference parameter, providing an accurate and reliable aging accumulation metric for the prediction model. Furthermore, collecting cycle counts at the full charge stage ensures standardization and consistency of data statistics. Compared to counting cycles at any point during the charging process, the full charge stage provides clear state determination criteria and stable electrochemical conditions, eliminating counting errors caused by external factors such as charging interruptions and power fluctuations.

[0157] Based on the above embodiment, as an optional embodiment, when the charging stage is a constant voltage charging stage, the constant voltage charging time and constant voltage charging capacity of the battery are collected;

[0158] The constant-voltage charging time is used to characterize the change in the polarization degree of the battery. The growth trend of the constant-voltage charging time of the battery with the number of cycles is extracted as a long-term feature through the parallel structure; the change of the constant-voltage charging time between adjacent cycles is extracted as a short-term feature through the parallel structure.

[0159] The constant-voltage charging capacity is used to characterize the degree of active material loss in the battery. The attenuation trend of the constant-voltage charging capacity is extracted as a long-term feature through the parallel structure, and the change of the constant-voltage charging capacity within a single cycle is extracted as a short-term feature through the parallel structure.

[0160] The constant voltage charging time refers to the length of time during the charging process when the charging voltage reaches the set charging cut-off voltage and the charging current continues to drop to the preset cut-off current value. In the disclosed embodiments, this can be understood as a time characteristic parameter reflecting the change in charge transfer impedance and the evolution of polarization within the battery. By quantifying the duration of the constant voltage charging phase, it indirectly characterizes the activity of the electrochemical reaction within the battery and the resistance to ion transfer. It is used to evaluate the impact of internal mechanism changes such as the thickening of the passivation layer on the surface of the electrode material and the intensification of electrolyte concentration polarization on the charging efficiency during battery aging.

[0161] Similarly, constant voltage charging capacity refers to the amount of electricity accumulated by the battery during the constant voltage charging phase from the start of charging to the point where the charging cutoff condition is reached. In the disclosed embodiments, this can be understood as a capacity characteristic parameter that characterizes the ability of the active material in the battery to participate in electrochemical reactions under constant voltage charging conditions. By quantifying the actual amount of electricity charged during the constant voltage phase, it directly reflects the available number of active lithium ions in the battery and the electrochemical activity level of the electrode material. This is used to evaluate the impact of capacity decay mechanisms such as loss of positive and negative electrode active materials and irreversible consumption of lithium ions on the battery's energy storage performance during battery aging.

[0162] Furthermore, during the battery charging process, the constant voltage charging stage is a key stage when the battery is close to a fully charged state. At this time, the electrochemical reaction inside the battery gradually tends to equilibrium, and the charging current gradually decays from a higher value to the cut-off current. The above process directly reflects the internal impedance characteristics of the battery and the state of the active substances.

[0163] Specifically, as the number of battery cycles increases, the solid electrolyte interface film on the negative electrode surface gradually thickens, the structural stability of the positive electrode active material decreases, and the ion conductivity of the electrolyte decays. These factors together lead to an increase in the internal impedance of the battery. The direct manifestation of the increased impedance is that during the constant voltage charging stage, the current decay rate slows down under the same voltage drive, and it takes longer to reach the charge cutoff current. Therefore, the constant voltage charging time tends to increase with the number of cycles. This change trend has good monotonicity and sensitivity, making the constant voltage charging time an effective indicator of the degree of battery aging.

[0164] At the same time, the change in constant voltage charging capacity directly reflects the attenuation process of the battery's energy storage capacity. During the aging process of the battery, the lithium deintercalation ability of the positive electrode material decreases, the lithium intercalation capacity of the negative electrode material decreases, and some lithium ions are permanently lost due to side reactions. The above factors cause the actual rechargeable capacity of the battery to gradually decrease. The constant voltage charging stage is an important part of battery capacity replenishment, and its capacity change can sensitively reflect the degree of loss of active substances. By monitoring the attenuation trend of the constant voltage charging capacity, the degree of degradation of the battery's current energy storage capacity relative to the initial state can be accurately assessed.

[0165] It is worth noting that the operating conditions during the constant-voltage charging phase are relatively standardized, with a fixed charging voltage and clear termination conditions, making the measurement results between different cycles highly comparable. In addition, the constant-voltage charging phase typically lasts for a moderate duration throughout the charging process, providing an ample data sampling window.

[0166] Figure 12 A characteristic evolution diagram of a battery constant voltage charging time as a function of cycle number provided in an embodiment of the present disclosure is schematically shown.

[0167] like Figure 12 As shown in the figure, during constant-voltage charging, the charging current continuously decreases until it reaches the preset cutoff current value. As the number of cycles increases, the constant-voltage charging time shows a trend of continuous growth. The constant-voltage charging time of the initial cycle (Cycle 1) is relatively short, marked as A1-1; however, after 300 cycles (Cycle 300), the constant-voltage charging time is significantly extended, marked as A1-300. The physical mechanism of this time extension phenomenon is that as the battery ages, the internal impedance of the battery gradually increases, including the combined increase of ohmic impedance, charge transfer impedance, and diffusion impedance. This results in a slower current decay rate under the same constant-voltage conditions, and it takes longer to reach the charging cutoff condition.

[0168] Therefore, the constant voltage charging time has good monotonicity and sensitivity, and can accurately reflect the changing trend of the battery's internal electrochemical impedance. The change from A1-1 to A1-300 reflects the gradual increase in the polarization degree and the gradual decrease in charging efficiency during battery use.

[0169] Figure 13 The figure schematically shows a capacity decay characteristic diagram of a battery constant voltage charging capacity as a function of cycle number provided by an embodiment of the present disclosure.

[0170] like Figure 13 As shown in the figure, in the initial stage (the first cycle), the constant voltage charging capacity is A2-1; after 100 cycles, the constant voltage charging capacity changes to A2-100; when aging continues to 200 cycles, the capacity further changes to A2-200; when the battery completes 300 cycles, the constant voltage charging capacity reaches A2-300.

[0171] The above capacity changes reflect the loss of active materials during battery aging. During battery use, the positive electrode material's ability to deintercalate and deintercalate lithium gradually decreases, the SEI film on the negative electrode material's surface continues to thicken, consuming active lithium, and lithium ions in the electrolyte are partially and irreversibly lost due to side reactions. These factors together cause the battery's actual chargeable capacity to change during constant-voltage charging. Changes in constant-voltage charging capacity directly reflect the availability of active materials within the battery and the activity level of the electrochemical reactions.

[0172] It's worth noting that the pattern of change in constant-voltage charging capacity is closely related to the battery's specific material system and aging mechanism. Different battery types may exhibit different capacity change trends, but constant-voltage charging capacity, as a key parameter characterizing the state of active materials, consistently provides an effective indicator of the battery's inherent health. Therefore, constant-voltage charging capacity can serve as an indirect health factor to measure battery aging, denoted as A2.

[0173] In one feasible implementation, electronic devices can automatically collect and monitor parameters during the constant-voltage charging phase through an integrated battery management system. Specifically, the battery management system continuously monitors the charging voltage and current during the charging process, using an algorithm to determine the switching point between charging phases. When the system detects that the charging voltage reaches a preset constant-voltage charging threshold, it automatically marks the start of the constant-voltage charging phase and initiates a dedicated parameter collection program.

[0174] During the constant-voltage charging phase, the battery management system continuously records the charging current value at a fixed sampling frequency through the current sensor, while also recording the corresponding timestamp information. The system calculates the rate of change of the charging current in real time and automatically marks the end of the constant-voltage charging phase when it detects that the charging current has dropped to the preset cutoff current threshold.

[0175] Based on the start and end times of the constant-voltage charging phase, the system automatically calculates the constant-voltage charging time, which is the time difference between the end and start times. To calculate the constant-voltage charging capacity, the system can use a current integration method to numerically integrate the current values ​​collected during the constant-voltage charging phase over time to obtain the cumulative charging capacity for that phase.

[0176] By using the disclosed embodiments, by collecting constant-voltage charging time and constant-voltage charging capacity during the constant-voltage charging phase, key characteristic parameters reflecting changes in the battery's internal electrochemical state can be obtained. Monitoring the constant-voltage charging time enables the prediction model to accurately assess the changing trends of the battery's internal impedance and the evolution of its polarization degree, while tracking the constant-voltage charging capacity directly reflects the decay of the battery's energy storage capacity.

[0177] Based on the above embodiment, as an optional embodiment, when the charging stage is a fully charged static stage, the battery's shelf start voltage and shelf end voltage are collected;

[0178] When the charging stage is the discharge start stage, the discharge start voltage of the battery is collected;

[0179] The shelf starting voltage is used to characterize the self-discharge characteristics of the battery. The changing trend of the shelf starting voltage with the number of cycles is extracted as a long-term feature through a parallel structure, and the difference in the shelf starting voltage between adjacent cycles is extracted as a short-term feature through a parallel structure.

[0180] The shelf termination voltage is used to characterize the polarization relaxation degree of the battery. The long-term decay trend of the shelf termination voltage is extracted as a long-term feature through the parallel structure, and the change of the shelf termination voltage during the static process is extracted as a short-term feature through the parallel structure.

[0181] The discharge start voltage is used to characterize the change in the internal resistance of the battery. The downward trend of the discharge start voltage with the aging process is extracted as a long-term feature through the parallel structure, and the sudden change of the discharge start voltage at the moment of discharge is extracted as a short-term feature through the parallel structure.

[0182] The shelf start voltage refers to the terminal voltage measured when the battery enters a static state immediately after charging is completed, that is, the battery voltage value at the moment charging is terminated. In the disclosed embodiments, this can be understood as a voltage characteristic parameter that reflects the internal electrochemical balance and the initial conditions of self-discharge of the battery when fully charged. By quantifying the battery voltage level at the end of charging, the internal polarization state and electrochemical potential distribution of the battery are characterized, and it is used to assess the starting baseline of the battery's self-discharge behavior during the static process and the degree to which the internal impedance affects the charge termination voltage.

[0183] Similarly, the shelf termination voltage refers to the stable terminal voltage value measured at the end of the resting phase of the battery, that is, the voltage value when the electrochemical reaction inside the battery tends to a state of equilibrium after a certain period of rest. In the embodiments of the present disclosure, it can be understood as a voltage characteristic parameter that characterizes the degree of polarization relaxation and electrochemical equilibrium recovery ability of the battery during the resting process. By quantifying the battery voltage level at the end of the resting phase, it reflects the degree of dissipation of concentration polarization inside the battery and the equilibrium state of the electrochemical reaction on the electrode surface, and is used to evaluate the influence of the internal ion transfer impedance of the battery and the activity of the electrode material on the voltage relaxation process.

[0184] Similarly, the discharge start voltage refers to the terminal voltage measured at the moment the battery begins discharging after the rest phase, i.e., the open-circuit voltage of the battery before the discharge current is applied. In the disclosed embodiments, this voltage characteristic parameter characterizes the battery's internal electrochemical state and load response characteristics at the start of discharge. By quantifying the battery voltage level at the moment of discharge initiation, it reflects the impact of the battery's internal ohmic impedance and polarization impedance on the terminal voltage, and is used to assess the battery's voltage retention capability under load conditions and the degree to which changes in internal resistance affect discharge performance.

[0185] Furthermore, the full charge and rest phase is a critical period for the battery to transition from a charged state to a balanced state. The voltage changes in this phase directly reflect the dynamic evolution of various electrochemical processes within the battery. During the charging process, there are various polarization phenomena inside the battery, and the polarization phenomenon makes the terminal voltage during the charging process higher than the actual open circuit voltage of the battery. When charging is completed and the battery enters the rest phase, the shelf start voltage retains the polarization information accumulated during the charging process. As the battery aging process progresses, the internal impedance of the battery increases and the polarization phenomenon intensifies, causing the shelf start voltage to show a specific change trend. By monitoring the long-term changes in the shelf start voltage with the number of cycles, the impact of battery aging on the charging polarization behavior can be accurately evaluated.

[0186] At the same time, during the static process, the various polarization effects inside the battery gradually dissipate, the concentration polarization weakens due to ion diffusion, the electrochemical polarization is eliminated due to reaction equilibrium, and the terminal voltage gradually converges to the true open circuit voltage. The shelf termination voltage reflects the final result of the polarization relaxation of the battery within a given static time, and its difference with the shelf starting voltage directly characterizes the degree of polarization relaxation of the battery. As the battery ages, the activity of the electrode material decreases, the ion transfer resistance increases, and the polarization relaxation process becomes slow and incomplete, resulting in a specific attenuation trend of the shelf termination voltage. By comparing the shelf termination voltages of different cycle periods, the impact of battery aging on the polarization relaxation ability can be effectively evaluated.

[0187] Correspondingly, after the rest phase, the battery has reached a relatively stable electrochemical equilibrium state, and the measured voltage at this time is close to the true open-circuit voltage of the battery. When the discharge current begins to be applied, the battery terminal voltage will immediately drop due to the internal resistance voltage drop. The difference between the discharge start voltage and the rest end voltage directly reflects the instantaneous internal resistance characteristics of the battery. As the battery ages, the internal resistance gradually increases, resulting in an increase in the decrease in the discharge start voltage relative to the rest voltage, and the absolute value of the discharge start voltage shows a downward trend. By monitoring the long-term change pattern of the discharge start voltage, the evolution of the battery's internal resistance can be accurately tracked.

[0188] At the same time, the instantaneous mutation characteristics of the discharge start voltage can sensitively reflect short-term changes in the battery state. Under normal circumstances, the magnitude of the discharge start voltage mutation is relatively stable. However, when the battery is abnormally aged or has an internal fault, the rapid change in internal resistance can cause the magnitude of the discharge start voltage mutation to increase abnormally. By monitoring the mutation characteristics of the discharge start voltage at the moment of discharge, abnormal changes in the battery state can be identified in a timely manner.

[0189] Figure 14 The figure schematically shows a voltage change characteristic curve of a battery provided by an embodiment of the present disclosure during a discharge process after full charge.

[0190] like Figure 14 As shown in the figure, it shows the complete voltage change process of the battery from full charge state to static stage and discharge start stage in different cycle periods. Figure 14 Three key voltage characteristic points can be observed from the voltage change curve: the starting voltage A3 after 15 minutes of storage after full charge, the ending voltage A4 after 15 minutes of storage after full charge, and the starting voltage A5 of constant current discharge.

[0191] When the battery enters the rest stage immediately after charging, the battery voltage will show a significant drop (A3). The drop in battery voltage is an electrochemical phenomenon caused by the self-discharge effect after charging is completed.

[0192] from Figure 14 Comparison of different cycles shows that the voltage drop at point A3 shows a changing trend as the number of cycles increases. Therefore, the voltage drop value at the initial stage of storage after full charge can be used as an indirect health factor A3 to measure the degree of battery aging.

[0193] like Figure 14 As shown in Figure 4, after 15 minutes of rest, the battery voltage tends to a relatively stable state (A4). Point A4 represents the final result of the polarization relaxation process of the battery within a given rest time.

[0194] from Figure 14 From the curve changes, it can be observed that during the static process, the battery voltage gradually decreases from point A3 and tends to stabilize, eventually reaching the voltage level of point A4. By comparing the voltage values ​​of point A4 at different cycle periods, it can be found that as the degree of battery aging deepens, the voltage level of point A4 shows a specific change trend. Therefore, the termination voltage A4 after 15 minutes of storage after full charge can be used as a parameter to characterize the degree of battery polarization relaxation. This parameter effectively reflects the changing law of the battery's internal electrochemical balance recovery ability and polarization dissipation efficiency by quantifying the voltage stability value of the battery after the standard static time.

[0195] like Figure 14As shown in Figure 5, when the battery completes the rest phase and enters the discharge phase, the battery voltage undergoes a significant sudden drop (A5) at the moment the constant discharge current is applied. This sudden drop in voltage directly reflects the degree to which the battery's internal impedance affects the load response.

[0196] from Figure 14 Comparing point A5 across different cycles clearly shows that the voltage drop at point A5 shows a changing trend as the battery ages. Therefore, the starting voltage A5 of constant-current discharge can be used as an indirect health factor to measure the extent of battery aging. This parameter quantifies the voltage spike at the start of discharge, effectively reflecting the impact of changes in the battery's internal impedance on its load response capability.

[0197] It is noteworthy that the three voltage characteristic parameters A3, A4, and A5 form a complete chain of voltage changes in the time series, reflecting the different electrochemical states of the battery at the initial rest period, the rest stabilization period, and the discharge start-up period, respectively. The combined analysis of these three parameters can provide comprehensive information on the multi-dimensional impedance characteristics within the battery: A3 reflects the dissipation process of charge polarization, A4 reflects the degree of recovery of rest equilibrium, and A5 reflects the impedance characteristics of the load response. Through the comprehensive monitoring and analysis of these three parameters, the predictive model can understand the aging mechanism and health status changes of the battery from multiple perspectives.

[0198] In a feasible implementation, the electronic device can automatically collect the above voltage parameters through a battery management system.

[0199] During the shelving start voltage acquisition process, the battery management system immediately stops charging and quickly switches to high-impedance measurement mode upon detecting that the charging current has dropped to the cutoff current threshold. Using a high-precision voltage sensor, the system completes the shelving start voltage measurement within one second of charging stopping, ensuring that the actual voltage state is captured at the moment charging ends.

[0200] During the rest phase, the battery management system continuously monitors the battery terminal voltage at regular intervals. The system uses an algorithm to analyze the voltage curve and identify the point at which the voltage stabilizes. If the voltage change over multiple consecutive measurements is less than a preset threshold, the system determines that the battery has reached rest equilibrium and automatically records the voltage at that point as the rest termination voltage.

[0201] During the discharge start voltage acquisition process, the battery management system prepares to initiate the discharge process after the rest phase. The system first measures the battery's open-circuit voltage before the discharge current is applied, which serves as a reference for the discharge start voltage. The system then rapidly applies a preset discharge current and, shortly after the current stabilizes, measures the battery's load voltage, recording this as the discharge start voltage.

[0202] By using the disclosed embodiments, key voltage characteristic parameters reflecting the multi-dimensional electrochemical properties of the battery can be obtained by collecting the shelf start voltage, shelf end voltage, and discharge start voltage during the full rest phase and the discharge start phase. These parameters comprehensively characterize the battery's aging state from three perspectives: self-discharge characteristics, polarization relaxation, and internal resistance change, providing rich characteristic information for the prediction model.

[0203] Figure 15 A battery health prediction feature importance analysis diagram based on a GRU model provided in an embodiment of the present disclosure is schematically shown.

[0204] like Figure 15 As shown in the figure, the horizontal axis represents six detection parameters, namely constant voltage charging time A1, constant voltage charging capacity A2, full charge shelf start voltage A3, full charge shelf end voltage A4, constant current discharge start voltage A5 and cycle number A6; the vertical axis represents the feature importance value, and the higher the value, the greater the contribution of the parameter in the battery health prediction process.

[0205] Depend on Figure 15 It can be seen that the GRU gating mechanism pays varying degrees of attention to each of the six input parameters. The top three parameters in importance are A1 > A6 > A4. This result is highly consistent with the predictions of the Transformer model and the actual physical and chemical mechanism analysis.

[0206] Among them, A1 has the highest feature importance score. This is because the constant voltage charging time directly reflects the change in the battery's internal impedance. As the battery ages, factors such as the thickening of the SEI film and the decrease in electrode material activity increase the internal resistance, which prolongs the constant voltage charging time. This parameter has good monotonicity and sensitivity.

[0207] A6 provides the model with a clear timing benchmark and cumulative aging measurement, directly quantifies the historical usage intensity of the battery, and provides a reference framework in the time dimension for changes in other parameters; A4 reflects the battery's polarization relaxation degree and electrochemical equilibrium recovery ability, effectively characterizing changes in the battery's internal impedance characteristics.

[0208] The following takes the constant voltage charging time A1 as an example to illustrate the specific role of the GRU gating mechanism.

[0209] The reset gate's information-focusing function: When the battery's internal polarization changes due to certain local factors, resulting in abnormal values ​​in A1, the reset gate can identify the abnormal change and quickly adjust the GRU's attention allocation mechanism. The reset gate weights the input parameters using a built-in weight matrix. When an abnormal change in A1 is detected, the reset gate reduces its reliance on historical hidden states, allowing the GRU to "focus" on the abnormal information in A1 at the current time step.

[0210] The update gate's information retention and filtering function: Because constant-voltage charging time increases monotonically with battery aging, historical information about battery aging accumulates continuously. The update gate dynamically determines how much historical information about A1 to retain in the current hidden state by comprehensively considering all input parameters. When A1 changes relatively slowly over time, indicating stable battery aging, the update gate's output value approaches 1, indicating that the model should retain a large amount of historical information about A1.

[0211] The GRU's gating mechanism also has the ability to adaptively learn changes in battery parameters. As battery cycles increase, the battery's physical and chemical properties gradually change. These changes are reflected in the numerical changes and evolving relationships between the six detection parameters. The GRU's gating mechanism can adaptively adjust its parameter learning strategy through dynamic adjustments to the weight matrix.

[0212] This adaptive capability ensures the prediction model maintains accurate performance across the different stages of battery aging. Whether the battery is in a stable period of slow aging or entering a late stage of rapid decay, the GRU's gating mechanism dynamically adjusts the model's learning focus and information processing based on the current parameter changes, thereby achieving continuous and accurate prediction of battery health.

[0213] The present disclosure also discloses an electronic device, including:

[0214] A collection device, used to collect multiple parameters of the battery at different charging stages to obtain detection parameters;

[0215] A processor is configured to obtain detection parameters of a battery; input the detection parameters into a prediction model to obtain long-term and short-term characteristics of the detection parameters, wherein the long-term characteristics are used to characterize aging information of the battery in a first time period, and the short-term characteristics are used to characterize aging information of the battery in a second time period, and the first time period is much longer than the second time period; and obtain prediction parameters of the battery based on the long-term and short-term characteristics; the prediction parameters are used to characterize the health status of the battery.

[0216] Figure 16 The block diagram of an electronic device provided by an embodiment of the present disclosure is schematically shown. Figure 16 The electronic device shown schematically is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0217] like Figure 16As shown, the electronic device 1600 according to an embodiment of the present disclosure includes an acquisition device (not shown in the figure) and a processor 1601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1602 or a program loaded from a memory 1608 into a random access memory (RAM) 1603. The processor 1601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1601 may also include onboard memory for caching purposes. The processor 1601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to the embodiment of the present disclosure.

[0218] Various programs and data required for the operation of the electronic device 1600 are stored in the RAM 1603. The processor 1601, the ROM 1602, and the RAM 1603 are connected to each other via a bus 1604. The processor 1601 performs various operations of the method flow according to the embodiment of the present disclosure by executing the programs in the ROM 1602 and / or the RAM 1603. It should be noted that the programs may also be stored in one or more memories other than the ROM 1602 and the RAM 1603. The processor 1601 may also perform various operations of the method flow according to the embodiment of the present disclosure by executing the programs stored in the one or more memories.

[0219] According to an embodiment of the present disclosure, electronic device 1600 may further include an input / output (I / O) interface 1604, which is also connected to bus 1604. System 1600 may also include one or more of the following components connected to I / O interface 1604: an input device 1606 including a keyboard, mouse, etc.; an output device 1607 including a cathode ray tube (CRT), liquid crystal display (LCD), display screen, etc., and speakers; a memory 1608 including a hard disk; and a communication unit 1609 including a network interface card such as a LAN card or modem. Communication unit 1609 performs communication processing via a network such as the Internet. A drive 1610 is also connected to I / O interface 1604 as needed. Removable media 1611, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 1610 as needed, so that computer programs read from the removable media can be installed into memory 1608 as needed.

[0220] According to an embodiment of the present disclosure, the method flow according to an embodiment of the present disclosure can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1609, and / or installed from the removable medium 1611. When the computer program is executed by the processor 1601, the above-mentioned functions defined in the system of the embodiment of the present disclosure are executed. According to an embodiment of the present disclosure, the system, device, apparatus, module, unit, etc. described above can be implemented by a computer program module.

[0221] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when executed, implements the method according to the embodiments of the present disclosure.

[0222] According to embodiments of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium. Examples include, but are not limited to, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0223] For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the ROM 1602 and / or the RAM 1603 described above and / or one or more memories other than the ROM 1602 and the RAM 1603 .

[0224] An embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method provided by the embodiment of the present disclosure. When the computer program product runs on an electronic device, the program code is used to enable the electronic device to implement the method provided by the embodiment of the present disclosure.

[0225] When the computer program is executed by the processor 1601, the above functions defined in the system / device of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0226] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal over a network medium, downloaded and installed via the communication portion 1609, and / or installed from removable media 1611. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0227] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the boxes may occur in an order different from that marked in the accompanying drawings. For example, two boxes shown in succession may actually be executed substantially in parallel, or they may sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, as well as the combination of boxes in the block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or may be implemented using a combination of dedicated hardware and computer instructions. It will be understood by those skilled in the art that the features described in the various embodiments and / or claims of the present disclosure may be combined and / or coupled in various ways, even if such combinations or couplings are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments and / or claims of the present disclosure may be combined and / or coupled in various ways, and all such combinations and / or couplings fall within the scope of the present disclosure.

[0228] The embodiments of the present disclosure are described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be used in combination to advantage. The scope of the present disclosure is defined by the appended claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A battery monitoring method, comprising: Obtain battery detection parameters; The detection parameters are obtained by collecting multiple charging parameters of the battery at different charging stages; Inputting the detection parameters into a prediction model to obtain long-term and short-term features of the detection parameters, wherein the long-term features are used to characterize aging information of the battery in a first time period, and the short-term features are used to characterize aging information of the battery in a second time period, wherein the first time period is much longer than the second time period; A prediction parameter of the battery is obtained based on the long-term characteristics and the short-term characteristics; the prediction parameter is used to characterize the health state of the battery.

2. The method according to claim 1, wherein the prediction model comprises a parallel structure, wherein the parallel structure is used to extract the long-term features and the short-term features of the detection parameters, and output a fusion feature obtained by fusing the long-term features and the short-term features; The prediction parameter is obtained based on the fusion feature.

3. According to the method of claim 2, the prediction model includes a series structure connected to the parallel structure, and the series structure is used to smooth the fusion features to obtain target features; the target features are used to perform regression prediction to obtain prediction parameters.

4. According to the method of claim 2, the parallel structure includes a first self-attention layer and a first gated recurrent unit, the first self-attention layer is used to extract the long-term features; the first gated recurrent unit is used to extract the short-term features; the series structure includes a second gated recurrent unit, and the second gated unit is connected in series to process the fusion features.

5. The method according to claim 1, wherein obtaining the detection parameters of the battery comprises: When the charging stage is a full-charge stage, the number of cycles of the battery at the time of full charge is collected. The number of cycles represents the cumulative degree of aging of the battery and is used to provide a timing benchmark for the prediction model.

6. The method according to claim 2, wherein obtaining the detection parameters of the battery comprises: When the charging stage is a constant voltage charging stage, collecting the constant voltage charging time and constant voltage charging capacity of the battery; The constant-voltage charging time is used to characterize the change in the polarization degree of the battery. The growth trend of the constant-voltage charging time of the battery with the number of cycles is extracted as the long-term feature through the parallel structure; the change of the constant-voltage charging time between adjacent cycles is extracted as the short-term feature through the parallel structure. The constant-voltage charging capacity is used to characterize the degree of active material loss of the battery. The attenuation trend of the constant-voltage charging capacity is extracted as the long-term feature through the parallel structure, and the change of the constant-voltage charging capacity within a single cycle is extracted as the short-term feature through the parallel structure.

7. The method according to claim 2, wherein obtaining the detection parameters of the battery comprises: When the charging stage is a full-charge rest stage, collecting the rest start voltage and rest end voltage of the battery; When the charging stage is a discharge start stage, collecting the discharge start voltage of the battery; The shelf starting voltage is used to characterize the self-discharge characteristics of the battery. The variation trend of the shelf starting voltage with the number of cycles is extracted as the long-term feature through the parallel structure, and the difference of the shelf starting voltage between adjacent cycles is extracted as the short-term feature through the parallel structure. The shelf termination voltage is used to characterize the polarization relaxation degree of the battery. The long-term decay trend of the shelf termination voltage is extracted as the long-term feature through the parallel structure, and the change of the shelf termination voltage during the static process is extracted as the short-term feature through the parallel structure. The discharge start voltage is used to characterize the change in the internal resistance of the battery. The downward trend of the discharge start voltage with the aging process is extracted as the long-term feature through the parallel structure, and the sudden change of the discharge start voltage at the moment of discharge is extracted as the short-term feature through the parallel structure.

8. The method according to claim 1, further comprising: generating a characteristic matrix based on the detection parameters, wherein the characteristic matrix is ​​used to characterize the battery state of the battery at a previous moment and a current moment; Position encoding is performed on each sequence in the feature matrix, and the encoded feature matrix is ​​input into a prediction model to obtain the long-term features and short-term features of the detection parameters.

9. A computer-readable storage medium having a computer program or instructions stored thereon, wherein the computer program or instructions, when executed by a processor, implements the steps of the method according to any one of claims 1 to 8.

10. An electronic device comprising: A collection device, used to collect multiple parameters of the battery at different charging stages to obtain detection parameters; A processor, configured to obtain detection parameters of the battery; The detection parameters are input into a prediction model to obtain long-term and short-term characteristics of the detection parameters, wherein the long-term characteristics are used to characterize aging information of the battery in a first time period, and the short-term characteristics are used to characterize aging information of the battery in a second time period, and the first time period is much longer than the second time period; prediction parameters of the battery are obtained based on the long-term and short-term characteristics; the prediction parameters are used to characterize the health status of the battery.

Images