Lithium battery residual capacity online detection method and device

Abstract

The invention discloses a lithium battery residual capacity online detection method and device, and the method comprises the steps: obtaining a terminal voltage signal, a charging and discharging current signal and a temperature signal of a battery, and carrying out the synchronous sampling and fusion of the terminal voltage and the charging and discharging current signal to form a voltage-current combined feature; extracting a polarization voltage component based on the joint feature and the temperature signal to perform battery equivalent circuit parameter identification, and estimating an initial value of a state of charge by using circuit parameters; performing state recursive filtering by using the initial value, generating a state correction and an estimation residual sequence according to a predicted value and terminal voltage deviation, performing iterative correction to form a real-time charge state, and generating an estimation confidence interval based on residual statistical characteristics; analyzing and identifying an aging drift component according to the residual sequence long-term accumulation characteristic, and extracting a parameter compensation factor to adaptively update a circuit parameter; and generating a residual capacity value according to the corrected circuit parameters and the real-time charge state, and outputting a detection result based on the residual capacity value and the estimation confidence interval to realize high-precision residual capacity detection.

Description

Technical Field

[0001] This invention relates to the field of new energy battery management technology, and in particular to a method and device for online detection of the remaining capacity of lithium batteries. Background Technology

[0002] As the core energy carrier for electric vehicles and energy storage devices, the accurate acquisition of the remaining capacity of lithium batteries directly affects the reliability of vehicle range prediction and the user experience. Inaccurate remaining capacity estimation can lead to range anxiety or unexpected breakdowns, and in severe cases, irreversible battery damage due to over-discharge. Existing capacity estimation techniques mainly rely on the open-circuit voltage method or the ampere-hour integration method. The open-circuit voltage method requires the battery to be stationary for a long time to obtain an accurate reading, making it difficult to meet the real-time monitoring needs during vehicle operation. While the ampere-hour integration method can continuously track changes in state of charge, the accumulated error increases over time, and it cannot autonomously detect capacity decay caused by battery aging.

[0003] Furthermore, lithium batteries inevitably undergo performance degradation during use, manifested as increased internal impedance and decreased usable capacity. This degradation exhibits differentiated evolution patterns under varying temperatures and operating conditions. Traditional detection methods typically employ fixed model parameters for calculations, neglecting the dynamic changes in battery state over aging periods, leading to a gradual deterioration in detection accuracy over time. Therefore, achieving high-precision state-of-charge tracking under dynamic operating conditions and simultaneously sensing battery aging to correct the detection model has become a pressing technical challenge in the field of online lithium battery remaining capacity detection. Summary of the Invention

[0004] This invention discloses an online detection method and device for the remaining capacity of lithium batteries, aiming to solve the problems of insufficient accuracy in tracking the state of charge under dynamic operating conditions and model mismatch caused by battery aging. It forms a joint feature by synchronously sampling and fusing terminal voltage and charging / discharging current signals, extracts polarization voltage components from temperature signals for circuit parameter identification and initial state of charge estimation, uses state recursive filtering and iterative correction to form the real-time state of charge and generate an estimation confidence interval, achieves adaptive updating of circuit parameters based on long-term cumulative feature analysis of residual sequences, and finally outputs the remaining capacity detection result based on the corrected parameters, providing reliable support for energy management of lithium battery-powered equipment.

[0005] The first aspect of this invention proposes an online detection method for the remaining capacity of a lithium battery, comprising the following steps: The terminal voltage signal, charge / discharge current signal, and temperature signal of the battery are acquired, and the terminal voltage signal and the charge / discharge current signal are synchronously sampled and fused to form a voltage-current joint feature; Based on the voltage-current joint characteristics and the temperature signal, the polarization voltage component is extracted. The battery equivalent circuit parameters are identified using the polarization voltage component to obtain the circuit parameters. The initial state of charge is estimated using the circuit parameters to obtain the initial state of charge value. The initial state of charge is used to perform state recursive filtering to form a predicted state of charge. A state correction amount and an estimated residual sequence are generated based on the deviation between the predicted state of charge and the terminal voltage signal. The predicted state of charge is iteratively corrected using the state correction amount to form a real-time state of charge. An estimated confidence interval is generated based on the statistical characteristics of the estimated residual sequence. Long-term cumulative feature analysis is performed on the estimated residual sequence to identify aging drift components. Parameter compensation factors are extracted from the aging drift components, and the circuit parameters are adaptively updated using the parameter compensation factors to form corrected circuit parameters. The remaining capacity value is generated based on the corrected circuit parameters and the real-time state of charge, and the detection result is output based on the remaining capacity value and the estimated confidence interval.

[0006] A second aspect of the present invention provides an online detection device for the remaining capacity of a lithium battery, comprising: The signal fusion module is used to acquire the battery's terminal voltage signal, charge / discharge current signal, and temperature signal, and to synchronously sample and fuse the terminal voltage signal and the charge / discharge current signal to form a voltage-current joint feature; The parameter identification module is used to extract the polarization voltage component based on the voltage-current joint feature and the temperature signal, identify the battery equivalent circuit parameters through the polarization voltage component to obtain the circuit parameters, and use the circuit parameters to estimate the initial state of charge to obtain the initial value of the state of charge. The state filtering module is used to perform state recursive filtering using the initial state of charge to form a predicted state of charge, generate a state correction amount and an estimated residual sequence based on the deviation between the predicted state of charge and the terminal voltage signal, iteratively correct the predicted state of charge using the state correction amount to form a real-time state of charge, and generate an estimated confidence interval based on the statistical characteristics of the estimated residual sequence. The aging compensation module is used to perform long-term cumulative feature analysis on the estimated residual sequence to identify aging drift components, extract parameter compensation factors from the aging drift components, and use the parameter compensation factors to adaptively update the circuit parameters to form corrected circuit parameters. The capacity output module is used to generate a remaining capacity value based on the corrected circuit parameters and the real-time state of charge, and to output the detection result based on the remaining capacity value and the estimated confidence interval.

[0007] The beneficial effects of this invention are reflected in the following points: 1. By clock-aligning the terminal voltage signal and the charging / discharging current signal to eliminate the delay difference of the acquisition channel, the self-discharge characteristics of the static state are extracted from the synchronized signal, and the dynamic response is segmented according to the operating condition intensity. On this basis, the ohmic voltage drop and polarization voltage drop are separated by combining the current step response, and a temperature compensation model is introduced to eliminate the influence of environmental factors on polarization characteristics, thus realizing the accurate extraction of battery dynamic response characteristics and reliable identification of equivalent circuit parameters. 2. Extended Kalman filtering is used to recursively estimate the state of charge, and the prediction information of the state transition model is fused and corrected with the terminal voltage observation information. By performing sliding window statistical analysis on the estimated residual sequence, systematic bias is identified and the degree of random fluctuation is quantified. Based on this, an estimation confidence interval that dynamically adjusts with the operating condition is generated, so that the detection result not only provides a point estimate but also a reliability boundary. 3. Finally, the weak aging drift signal is amplified by multi-cycle cumulative summation of the estimated residual sequence. The monotonic drift trend segment and aging acceleration characteristics are identified by segmented slope analysis. The drift slope parameter is converted into capacity correction coefficient and internal resistance correction coefficient to realize the adaptive update of circuit model parameters with battery aging state. The actual usable capacity is calculated based on the corrected model parameters and the remaining capacity detection result is output in combination with real-time state of charge, ensuring the stability of detection accuracy throughout the battery's entire life cycle.

[0008] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0009] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0010] Figure 1 This is a flowchart illustrating an online detection method for the remaining capacity of a lithium battery according to the present invention.

[0011] Figure 2 This is a structural block diagram of an online detection device for the remaining capacity of a lithium battery according to the present invention. Detailed Implementation

[0012] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0013] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0014] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0015] The technical solutions of the embodiments of this application will be described below.

[0016] like Figure 1 As shown, this embodiment of the invention provides an online detection method for the remaining capacity of a lithium battery, including the following steps S110-S150: Step S110: Acquire the battery's terminal voltage signal, charge / discharge current signal, and temperature signal; synchronously sample and fuse the terminal voltage signal and charge / discharge current signal to form a voltage-current joint feature.

[0017] Specifically, the system acquires the battery's terminal voltage, charge / discharge current, and temperature signals. The terminal voltage signal is obtained from the positive and negative terminals of the battery via a differential voltage acquisition circuit, covering the typical operating voltage range of lithium batteries from 2.5V to 4.2V. The sampling frequency of the terminal voltage signal is set to 100Hz to capture the rapid dynamic response during charging and discharging. Each sampling point is accompanied by a timestamp, recording the precise moment of voltage sampling and providing a reference for subsequent multi-signal time alignment. The charge / discharge current signal is acquired from the battery's main circuit via a Hall effect current sensor, with a range set to ±100A, covering the high-rate charge / discharge conditions of electric vehicle power batteries. The sampling frequency of the charge / discharge current signal is consistent with that of the terminal voltage signal at 100Hz, providing a basis for synchronous analysis in the time domain. The consistency of the sampling frequency reduces the complexity of subsequent clock alignment processing. The charge / discharge current signal uses a bipolar representation, with positive values ​​representing discharge current and negative values ​​representing charging current. A dead zone of ±10mA is set near the zero value to eliminate the influence of sensor zero-point drift. Temperature signals are acquired through an NTC thermistor attached to the battery surface. The temperature measurement range covers -40℃ to 85℃, with a temperature resolution of 0.1℃. The sampling frequency is set to 1Hz. The temperature sampling frequency is lower than that of the voltage and current signals because the battery temperature changes relatively slowly, and its thermal time constant is usually on the order of tens of seconds to several minutes.

[0018] In some embodiments, the step of synchronously sampling and fusing the terminal voltage signal and the charging / discharging current signal to form a voltage-current joint feature includes: clock-aligning the terminal voltage signal and the charging / discharging current signal to obtain a synchronization signal pair; identifying weak current fluctuations in a static state from the synchronization signal pair and extracting self-discharge feature parameters; segmenting the synchronization signal pair based on the self-discharge feature parameters to form a segmented response feature; and constructing a voltage-current joint feature through the segmented response feature.

[0019] The terminal voltage signal and the charging / discharging current signal are clock-aligned to obtain a synchronization signal pair. Although the terminal voltage signal and the charging / discharging current signal use the same sampling frequency, there is an inherent delay difference of about 1-3ms between the two acquisition channels. Although this delay difference is small on a time scale, it may lead to misjudgment of the voltage-current timing relationship in dynamic response analysis, affecting the accurate identification of internal resistance and polarization characteristics. Clock alignment uses a step edge in the charging / discharging current signal with an amplitude change rate exceeding 10A / s as the reference event, searches for the corresponding voltage response start point in the terminal voltage signal, and calculates and compensates for the delay difference between the two signals through cross-correlation analysis and interpolation methods. The synchronization signal pair is generated after clock alignment and includes the aligned voltage value sequence, current value sequence, and a unified timestamp sequence. The uniformity of the timestamps ensures that the various feature parameters extracted from the synchronization signal pair have a strict correspondence in the time dimension. The number of valid step events identified in the charge and discharge current signal determines the reliability of clock alignment. When there are more than 20 valid step events in the sampling window, the alignment accuracy of the synchronization signal pair can reach ±0.3ms, which can meet the requirements of battery dynamic characteristic analysis.

[0020] Self-discharge characteristic parameters are extracted by identifying weak current fluctuations in the quiescent state from the synchronization signal pair. Determining the quiescent state requires setting both a current threshold and a duration condition. The battery is considered to have entered the quiescent state when the absolute value of the current in the synchronization signal pair remains below 50mA for more than 60 seconds. This 50mA threshold is set to five times the standard deviation of the current sensor's zero-point drift to improve the robustness of the determination. In the quiescent state, the current component in the synchronization signal pair is not strictly zero, but exhibits a weak DC bias with a mean of approximately 0.5-5mA superimposed with small random fluctuations. This weak current reflects the continuous self-discharge process inside the battery, originating from microcurrents caused by impurity ions in the electrolyte and slow chemical side reactions of the active materials. The voltage component in the quiescent segment of the synchronization signal pair shows a slow, monotonically decreasing trend. The voltage drop rate depends on the battery's self-discharge intensity and current state of charge. The voltage drop slope is obtained by least-squares linear fitting of this trend, with the slope unit being mV / h. The quiescent threshold in the self-discharge characteristic parameters is the larger of three times the standard deviation of the quiescent current and 50mA. If there are multiple discontinuous quiescent segments in the synchronization signal pair, the voltage drop slope in the self-discharge characteristic parameters is the weighted average of the fitting results for each segment, with the weight proportional to the duration of each segment. Longer quiescent segments contribute more to the estimation of the self-discharge rate. The self-discharge characteristic parameters include the voltage drop slope and the quiescent threshold.

[0021] Based on self-discharge characteristic parameters, the synchronization signal pair is segmented into operating conditions to form segmented response characteristics. The static condition threshold in the self-discharge characteristic parameters serves as the criterion for determining the operating condition boundary. Continuous segments in the synchronization signal pair with an absolute current value below this threshold are classified as static operating conditions, while continuous segments with an absolute current value above this threshold are classified as dynamic operating conditions. Operating condition segmentation involves time-domain cutting of the synchronization signal pair, with the cutting boundary located at the moment when the current amplitude crosses the static condition threshold. The boundary moment is precisely located using a linear interpolation method. The voltage drop slope in the self-discharge characteristic parameters is used to compensate for the influence of the static segment on the initial state of the dynamic segment. The starting voltage of the dynamic segment is increased by the corresponding static loss, which eliminates the interference of voltage drop caused by self-discharge during the static period on the dynamic response analysis. The segmented response characteristics extract four response parameters for each dynamic operating condition segment: peak-to-peak current change, voltage response amplitude, first response time constant, and intensity level. The intensity level of the synchronization signal in the dynamic operating condition segment is classified according to the ratio of the average current to the nominal battery capacity. A ratio below 0.3C indicates a low-rate operating condition, a ratio between 0.3C and 1C indicates a medium-rate operating condition, and a ratio above 1C indicates a high-rate operating condition. Different intensity levels correspond to different polarization characteristics and internal resistance performance. The segmented response characteristics record the end voltage value of the longest resting operating condition segment. This voltage value, after self-discharge compensation, serves as the basis for estimating the open-circuit voltage. The open-circuit voltage is a direct indicator of the battery's state of charge. The end voltage of the longest resting segment is closest to the true open-circuit voltage due to sufficient relaxation during the polarization process.

[0022] A voltage-current joint characteristic is constructed using segmented response characteristics. The ratio of the voltage response amplitude to the peak-to-peak current change in each dynamic operating condition segment of the segmented response characteristics is defined as the apparent impedance of that segment, calculated using the formula Z_app = ΔV / ΔI, where Z_app is the apparent impedance in mΩ, ΔV is the voltage response amplitude in mV, and ΔI is the peak-to-peak current change in A. The apparent impedance comprehensively reflects the superposition effect of ohmic internal resistance and polarization internal resistance. The voltage-current joint characteristic classifies and statistically analyzes the dynamic operating condition segments in the segmented response characteristics according to intensity levels, calculating the average apparent impedance for low-rate, medium-rate, and high-rate operating condition segments respectively. The three sets of apparent impedance values ​​obtained from the classification and statistics characterize the internal resistance level of the battery under different operating conditions. The time constants of each level of operating condition segments in the segmented response characteristics are also averaged. The decreasing time constant with increasing operating condition intensity reflects the dynamic characteristics of the polarization process; polarization establishment is faster under high-current conditions. The voltage-current joint feature also includes an open-circuit voltage estimate extracted from the segmented response features. This estimate is based on the end voltage of the longest resting segment after self-discharge compensation. The completeness of the dynamic operating condition segment coverage in the segmented response features determines the confidence level of the voltage-current joint feature. A high confidence level is achieved when all three intensity levels have at least three valid operating condition segments; the confidence level decreases accordingly when a segment of a certain intensity level is missing. The confidence level reflects the representativeness of the voltage-current joint feature to the battery's full operating condition characteristics. The voltage-current joint feature includes the apparent impedance and time constant values ​​for the three intensity levels, the open-circuit voltage estimate, and a confidence level indicator.

[0023] Step S120: Based on the voltage-current joint features and temperature signal, the polarization voltage component is extracted. The equivalent circuit parameters of the battery are identified by the polarization voltage component to obtain the circuit parameters. The initial state of charge is estimated using the circuit parameters to obtain the initial value of the state of charge.

[0024] In some embodiments, the step of extracting the polarization voltage component based on the voltage-current joint characteristics and the temperature signal includes: separating voltage components according to the voltage-current joint characteristics to obtain a dynamic response component; separating the ohmic voltage drop from the dynamic response component to obtain a pure polarization response; performing temperature compensation on the pure polarization response using the temperature signal to form a temperature-corrected polarization value; and using the temperature-corrected polarization value to quantify charge-discharge asymmetry to form a polarization voltage component.

[0025] The voltage component is separated based on the voltage-current joint characteristics to obtain the dynamic response components. The open-circuit voltage estimate in the voltage-current joint characteristics serves as the voltage reference. The difference between the sampled terminal voltage and this reference is defined as the total dynamic voltage offset, which includes both ohmic voltage drop and polarization voltage drop components. The three levels of apparent impedance values ​​in the voltage-current joint characteristics are used to estimate the contribution of ohmic voltage drop under different operating conditions. The ohmic voltage drop for low-rate operating conditions is calculated by multiplying the low-rate apparent impedance by the instantaneous current. For medium-rate and high-rate operating conditions, a similar calculation is performed using the corresponding level of apparent impedance. Voltage component separation decomposes the total dynamic voltage offset into an estimated ohmic voltage drop and the remaining dynamic components. The estimated ohmic voltage drop is calculated by automatically selecting the corresponding intensity level of apparent impedance value based on the instantaneous current amplitude. The three levels of time constant values ​​in the voltage-current joint characteristics are used to verify the rationality of the time-domain characteristics of the dynamic response components. An anomaly flag is triggered when the measured relaxation time of the dynamic response component deviates from the corresponding level of time constant by more than 30%. The confidence level in the voltage-current joint characteristics determines the parameter selection strategy for the voltage component separation process. At high confidence levels, the corresponding apparent impedance value is directly used; at low confidence levels, a weighted average of the three apparent impedance levels is used to improve robustness. The dynamic response component is defined as the difference between the total dynamic voltage offset and the ohmic voltage drop estimate. This component mainly includes the voltage response caused by electrochemical polarization and concentration polarization. The dynamic response component is output in time series form, with each sampling point in the series recording the polarization voltage offset value and the corresponding timestamp information at that moment.

[0026] For example, obtaining a pure polarization response by separating the ohmic voltage drop from the dynamic response components includes: identifying current step transient response characteristics based on the dynamic response components; extracting an ohmic voltage drop by extracting an instantaneous voltage drop from the current step transient response characteristics; generating an internal resistance characteristic parameter by performing a ratio correlation between the ohmic voltage drop and the current step transient response characteristics; and obtaining a pure polarization response by separating the voltage drop from the dynamic response components based on the internal resistance characteristic parameter.

[0027] The dynamic response component is used to identify the transient response characteristics of current step jumps. The dynamic response component exhibits a characteristic transient response pattern at the moment of a current step change. This pattern consists of two parts: a rapid jump phase and a slow relaxation phase. The rapid jump phase typically lasts less than 10 ms, while the slow relaxation phase lasts from several seconds to tens of seconds. The identification of current step jump transient response characteristics uses a current change rate exceeding 10 A / s as the trigger condition. This threshold effectively distinguishes between a true step event and a slowly gradual current change. Data windows of 50 ms are taken before and after the step trigger moment from the dynamic response component for analysis. The voltage waveform within each window constitutes a complete transient response record of a single step event, including two key parameters: the steady-state voltage before the step and the instantaneous voltage after the step. All step events in the dynamic response component that meet the trigger condition are processed one by one and summarized to form a current step jump transient response characteristic dataset. The current step transient response characteristics also record the current change corresponding to each step event. The current change is defined as the difference between the steady-state current after the step and the steady-state current before the step. A positive value indicates a step in the direction of current increase, and a negative value indicates a step in the direction of current decrease. The current step transient response characteristics classify step events into two groups according to the direction of current change: the charging direction group and the discharging direction group, which facilitates the separate handling of ohmic voltage drop differences during the charging and discharging processes.

[0028] Ohmic voltage drop is obtained by extracting the instantaneous voltage drop from the transient response characteristics of current step events. The difference between the instantaneous voltage after a step and the steady-state voltage before the step is defined as the instantaneous voltage drop of that step event. This instantaneous voltage drop occurs within a very short time after the current change and is mainly caused by the ohmic internal resistance, before the polarization process is fully established. The extraction of ohmic voltage drop involves calculating the instantaneous voltage drop for each step event in the current step transient response characteristics, and distinguishing between charging-direction and discharging-direction steps based on the sign of the current change. The absolute value of the instantaneous voltage drop for discharging-direction step events is taken, and then the arithmetic mean is calculated to obtain the average discharging ohmic voltage drop. This mean reflects the average voltage drop level caused by the ohmic internal resistance under discharging conditions. Charging-direction step events are processed using the same method to obtain the average charging ohmic voltage drop. The difference between the two reflects the asymmetry of the ohmic internal resistance in the charging and discharging directions. The ohmic voltage drop includes the average voltage drop in both the discharging and charging directions and the corresponding average current change. If the number of effective step events in the current step transient response characteristics is less than 5, the reliability rating of the ohmic voltage drop should be set to low. In this case, it is recommended to extend the data acquisition time to obtain more effective events.

[0029] Internal resistance characteristic parameters are generated by correlating the ohmic voltage drop with the current step transient response characteristics. The average discharge ohmic voltage drop in the ohmic voltage drop is divided by the average current change in the discharge direction in the current step transient response characteristics to obtain the estimated discharge ohmic internal resistance. The calculation formula is R_ohm = ΔV_ohm / ΔI, where R_ohm is the ohmic internal resistance in mΩ, ΔV_ohm is the average ohmic voltage drop in mV, and ΔI is the average current change in A. The charging ohmic internal resistance is calculated using the same method. The ratio of the charging ohmic internal resistance to the discharging ohmic internal resistance is defined as the charge-discharge internal resistance ratio. This ratio is typically between 0.85 and 1.15; deviations from this range indicate significant charge-discharge asymmetry in the battery. In the current step transient response characteristics, ohmic internal resistance estimates are calculated for each step event. The dispersion of all estimates is measured by the ratio of the standard deviation to the mean, defined as the internal resistance dispersion. An internal resistance dispersion exceeding 20% ​​indicates poor consistency of the ohmic voltage drop. The number of events in both the charging and discharging directions in the current step transient response characteristics jointly determines the statistical reliability of the internal resistance characteristic parameters; higher reliability is achieved when both groups have more than 10 events. The internal resistance characteristic parameters include discharging ohmic internal resistance, charging ohmic internal resistance, the charge-discharge internal resistance ratio, and internal resistance dispersion. The reliability of the ohmic voltage drop is transferred to the internal resistance characteristic parameters; low-reliability ohmic voltage drops correspond to low-confidence internal resistance characteristic parameters.

[0030] The pure polarization response is obtained by voltage drop separation based on the internal resistance characteristic parameter and the dynamic response component. The discharge ohmic internal resistance in the internal resistance characteristic parameter is multiplied by the forward current to obtain the reconstructed ohmic voltage drop value at each sampling time under discharge conditions. Under charging conditions, the charging ohmic internal resistance is used for corresponding calculations. The pure polarization response is obtained by subtracting the reconstructed ohmic voltage drop value at the corresponding time from the dynamic response component. This subtraction operation is performed one by one at each sampling point in the time series of the dynamic response component, and the result retains the original timestamp information. The charge / discharge internal resistance ratio in the internal resistance characteristic parameter is used to determine whether to use the discharge ohmic internal resistance or the charging ohmic internal resistance at the current time. The discharge ohmic internal resistance is used when the current is positive, and the charging ohmic internal resistance is used when the current is negative. The pure polarization response inherits the time series structure of the dynamic response component, with the sequence length and sampling interval remaining consistent. The amplitude of the pure polarization response is typically 60%-80% of the amplitude of the dynamic response component. The internal resistance dispersion in the internal resistance characteristic parameter affects the uncertainty estimation of the pure polarization response; the larger the dispersion, the wider the confidence interval of each sampling point of the pure polarization response. After the ohmic voltage drop component in the dynamic response is separated, the pure polarization response retains only the voltage response components caused by electrochemical polarization and concentration polarization. The pure polarization response exhibits exponential relaxation characteristics after a current step, and the relaxation time constant reflects the kinetic rate of the polarization process.

[0031] Temperature-corrected polarization values ​​are generated by temperature compensation of the pure polarization response using a temperature signal. The amplitude of the pure polarization response is significantly affected by temperature. At low temperatures, the amplitude increases due to a decrease in the electrochemical reaction rate and the ion diffusion coefficient, while at high temperatures, the amplitude decreases due to increased reactivity and reduced mass transfer resistance. The temperature signal is aligned with the pure polarization response by timestamp. Since the sampling frequency of the temperature signal is 1 Hz while that of the pure polarization response is 100 Hz, the temperature signal is extended to the same time resolution as the pure polarization response using a zero-order hold method. Temperature compensation employs the Arrhenius correction model, with the compensation formula V_pol_comp = V_pol × exp[E_a / R × (1 / T - 1 / T_ref)], where V_pol_comp is the temperature-corrected polarization voltage in mV, V_pol is the original value of the pure polarization response in mV, E_a is the apparent activation energy of the polarization process in J / mol, R is the gas constant (8.314 J / (mol·K), T is the absolute temperature scale value of the current temperature in K, and T_ref is the reference temperature (298 K). When the temperature signal is below 0℃, the compensation coefficient of the pure polarization response is greater than 1, indicating that the polarization response at low temperatures needs to be downward corrected to reflect the equivalent level at the standard temperature. When the temperature signal is above 25℃, the compensation coefficient is less than 1. The temperature-corrected polarization values ​​are output in time series format, with the sequence structure consistent with the pure polarization response. The polarization voltage at each sampling point has been normalized to the 25℃ reference temperature condition. The temperature signal serves as a correction benchmark throughout the compensation process, ensuring that data collected under different temperature conditions are comparable.

[0032] The charge-discharge asymmetry quantification is performed using temperature-corrected polarization values ​​to form the polarization voltage component. The temperature-corrected polarization values ​​exhibit different amplitude characteristics during charging and discharging, with the discharge polarization amplitude typically being 10%-30% greater than the charging polarization amplitude. This asymmetry stems from the difference in lithium-ion insertion / extraction kinetics between the positive and negative electrode materials, as well as the directional selectivity of the solid electrolyte interface film. The charge-discharge asymmetry quantification extracts the polarization amplitude statistics for the charging and discharging stages from the temperature-corrected polarization values. The charging polarization amplitude is taken as the arithmetic mean of the absolute values ​​of the temperature-corrected polarization values ​​under charging conditions, and the discharging polarization amplitude is processed using the same method for discharging data. The asymmetry factor of the polarization voltage component is defined as the ratio of the discharging polarization amplitude to the charging polarization amplitude. An asymmetry factor greater than 1 indicates more significant discharging polarization; the typical value range for this factor is 1.05 to 1.35. The time constant information in the temperature-corrected polarization value is extracted by performing a double-exponential fitting on the relaxation curve, yielding two parameters: the fast polarization time constant and the slow polarization time constant. The fast polarization time constant corresponds to the electrochemical polarization process, while the slow polarization time constant corresponds to the concentration polarization process. The voltage relaxation curve in the resting segment of the temperature-corrected polarization value is used to extract the open-circuit voltage characteristic curve, which reflects the equilibrium voltage level under different charging states. The polarization voltage component includes the polarization amplitude in both charging and discharging directions, the asymmetry factor, the fast and slow polarization time constants, and the open-circuit voltage characteristic curve.

[0033] The battery equivalent circuit parameters are obtained by identifying the polarization voltage component. The charging and discharging polarization amplitudes in the polarization voltage component reflect the intensity of the battery's polarization characteristics under different current directions. The average value of these two values ​​serves as the reference input for identifying the equivalent polarization resistance, which is equal to the average polarization amplitude divided by the reference current value. The battery equivalent circuit adopts a second-order RC model structure. The model includes an ideal voltage source representing the open-circuit voltage, a series resistor representing the ohmic internal resistance, and two sets of parallel RC networks representing the electrochemical polarization and concentration polarization processes, respectively. The fast polarization time constant in the polarization voltage component is directly assigned to the time constant parameter of the first set of RC networks. This time constant typically corresponds to the fast response process of electrochemical polarization in the range of 1-10 seconds. The slow polarization time constant in the polarization voltage component is directly assigned to the time constant parameter of the second set of RC networks. This time constant typically corresponds to the slow relaxation process of concentration polarization in the range of 30-300 seconds. The asymmetry factor in the polarization voltage component is used to configure the direction-dependent parameters of the equivalent circuit. The polarization resistance in the discharge direction is amplified proportionally by the asymmetry factor, while the polarization resistance in the charging direction is correspondingly reduced, enabling the model to characterize the asymmetric charging and discharging characteristics of the battery. The open-circuit voltage characteristic curve in the polarization voltage component is stored in the open-circuit voltage lookup table of the circuit parameters after piecewise linearization. The lookup table covers the 0%-100% state of charge range, with a resolution of 101 voltage sampling points corresponding to 1% state of charge. The circuit parameters include ohmic internal resistance, polarization resistance and time constant of two sets of RC networks, the open-circuit voltage lookup table, and parameter confidence indicators.

[0034] In some embodiments, the step of using the circuit parameters to estimate the initial state of charge and obtain an initial state of charge value includes: constructing a mapping relationship between open-circuit voltage and state of charge based on the circuit parameters; extracting internal resistance characteristic parameters from the circuit parameters to generate a polarization compensation factor; performing a coarse state estimation through the mapping relationship to obtain a coarse state of charge value; and using the polarization compensation factor to dynamically compensate the coarse state of charge value to obtain an initial state of charge value.

[0035] A mapping relationship between open-circuit voltage and state of charge (SOC) is constructed based on circuit parameters. An open-circuit voltage lookup table in the circuit parameters provides the basic correspondence between open-circuit voltage and SOC. This lookup table was obtained during battery factory calibration through low-current charge-discharge experiments, covering the complete SOC range of 0%-100%. The mapping relationship involves piecewise linearization of the open-circuit voltage lookup table in the circuit parameters, dividing the SOC range into 10 equal-width sub-intervals, each with a width of 10% of the SOC. Within each sub-interval, the open-circuit voltage and SOC have an approximately linear relationship. The SOC resolution of the lookup table in the circuit parameters is 1%, corresponding to 101 open-circuit voltage sampling points. After piecewise linearization, 11 endpoint voltage values ​​are retained for interpolation calculations. The mapping relationship uses a lookup table plus linear interpolation to achieve rapid SOC estimation. After inputting the open-circuit voltage value, a binary search is first used to locate the corresponding sub-interval, and then linear interpolation is performed within that sub-interval to obtain the estimated SOC value. The confidence level of the parameters in the circuit parameters affects the strategy for constructing the mapping relationship. When the confidence level is low, a conservative interpolation strategy is adopted to reduce the risk of estimation bias. The effective voltage range of the mapping relationship covers the minimum value of 2.8V to the maximum value of 4.2V of the open-circuit voltage curve in the circuit parameters. When the input voltage exceeds this range, the state of charge value of the corresponding boundary is returned.

[0036] The polarization compensation factor is generated by extracting internal resistance characteristic parameters from the circuit parameters. The first and second RC network time constants in the circuit parameters correspond to the response characteristics of electrochemical polarization and concentration polarization, respectively. The first RC network time constant is typically in the range of 1-10 seconds, and the second RC network time constant is typically in the range of 30-300 seconds. The polarization compensation factor is generated by reading the polarization resistance of the first and second RC networks from the circuit parameters; their sum is defined as the total polarization resistance, which serves as the benchmark for compensation calculation. The sum of the ohmic internal resistance and the total polarization resistance in the circuit parameters constitutes the total internal resistance of the battery. The total internal resistance determines the magnitude of the deviation of the terminal voltage from the off-circuit voltage under load conditions. The equivalent time constant in the polarization compensation factor is taken as the weighted average of the time constants of the two RC networks in the circuit parameters. The calculation formula is τ_eq=(R_p1×τ_1+R_p2×τ_2) / (R_p1+R_p2), where τ_eq is the equivalent time constant in seconds, R_p1 and R_p2 are the polarization resistances of the first and second RC networks in mΩ, and τ_1 and τ_2 are the time constants of the first and second RC networks in seconds. This weighting method assigns a higher weight to the RC network with a larger polarization resistance. If the confidence level of a certain group of RC network parameters is marked as low, the weight of that group of parameters in the polarization compensation factor calculation is reduced accordingly to reduce the impact of uncertainty. The polarization compensation factor comprehensively characterizes the amplitude and time-domain characteristics of the battery polarization process through the two parameters of total polarization resistance and equivalent time constant.

[0037] A rough estimate of the state of charge (SOC) is obtained through a mapping relationship. The mapping relationship uses the currently measured terminal voltage as the query input for a preliminary estimate of the SOC. The query process directly substitutes the terminal voltage into the mapping relationship for a table lookup operation. This rough estimation process, as a fast estimation method, does not consider the influence of polarization voltage drop. The rough SOC estimate is obtained by finding the SOC corresponding to the terminal voltage in the mapping relationship. The lookup process first uses a bisection method to quickly locate the sub-interval to which the terminal voltage belongs among the 11 terminal voltage values ​​in the mapping relationship. Then, within this sub-interval, linear interpolation is performed based on the voltage values ​​at both ends of the interval and the SOC value to obtain the estimated SOC value. The open-circuit voltage curve in the mapping relationship is relatively flat in the middle section of the SOC range of 30%-70%. In this section, a 10mV voltage change may correspond to a 5%-10% SOC change. The rough estimation accuracy is relatively low in the middle section and is sensitive to the accuracy of the voltage input. The mapping relationship shows a significantly steeper slope at the two ends of the state of charge (SCC) approaching 0% or 100%. Within these segments, the same voltage change corresponds to a smaller SCC change, resulting in relatively higher coarse estimation accuracy at these two ends. The estimation error of the coarse SCC estimate mainly originates from the deviation between the terminal voltage and the actual open-circuit voltage, as well as calibration errors and aging drift in the mapping relationship itself. This deviation will be eliminated through polarization correction in the subsequent dynamic compensation stage. The coarse SCC estimate is output as a percentage, with a numerical range limited to 0% to 100%.

[0038] The initial state of charge (SOC) is obtained by dynamically compensating the coarse estimate using a polarization compensation factor. The total polarization resistance in the polarization compensation factor is multiplied by the current current to obtain the steady-state polarization voltage drop estimate, which represents the final voltage drop level after the polarization process is fully established. The dynamic compensation of the coarse SOC estimate considers the transient characteristics of the polarization process; when the current changes, the polarization voltage needs several time constants to transition from the old steady state to the new steady state. The equivalent time constant in the polarization compensation factor is used to calculate the establishment ratio of the polarization voltage relative to the steady-state value at the current moment. The transient establishment ratio factor is calculated exponentially based on the ratio of the time elapsed after the current change to the equivalent time constant, with a value ranging from 0 to 1. The actual polarization voltage drop at the current moment is equal to the steady-state polarization voltage drop multiplied by the transient establishment ratio factor. This actual polarization voltage drop is used to correct the open-circuit voltage estimate corresponding to the coarse SOC estimate. After transient polarization compensation, the corrected open-circuit voltage value is obtained from the coarse SOC estimate. This corrected value is then substituted back into the mapping relationship for a second lookup to obtain the initial SOC value. The compensation effect of the polarization compensation factor is more significant under high current conditions. Dynamic compensation at 1C rate can reduce the state of charge estimation error by 2%-3%, while the compensation effect is relatively limited under low current conditions. The initial state of charge value includes the state estimate, the timestamp of the estimation time, and the estimation confidence level. The confidence level is comprehensively evaluated based on the reliability level and mapping relationship of each parameter of the polarization compensation factor at the local sensitivity of the current operating point.

[0039] Step S130: Use the initial state of charge value to perform state recursive filtering to form the predicted state of charge value. Generate a state correction amount and an estimated residual sequence based on the deviation between the predicted state of charge value and the terminal voltage signal. Iteratively correct the predicted state of charge value using the state correction amount to form the real-time state of charge. Generate an estimated confidence interval based on the statistical characteristics of the estimated residual sequence.

[0040] Specifically, the initial state of charge (SOC) is used for state recursive filtering to generate predicted SOC values. The estimated SOC values ​​from the initial SOC values ​​serve as the initial state input to the extended Kalman filter, which uses the battery SOC and polarization voltage as state variables to form a two-dimensional state vector. The estimated confidence level from the initial SOC values ​​is used to initialize the filter's state covariance matrix. High confidence corresponds to a smaller initial covariance value, indicating a higher degree of confidence in the initial value, while low confidence corresponds to a larger initial covariance value, allowing for greater adjustments to the filter in subsequent iterations. The state recursive filtering method constructs a state transition equation based on the ampere-hour integral method. The transition equation is SOC_k = SOC_{k-1} - η × I_k × Δt / C_n, where SOC_k is the predicted state of charge at time k (in %), SOC_{k-1} is the estimated state of charge at the previous time (in %), η is the Coulomb efficiency coefficient (0.98 for charging, 1.0 for discharging), I_k is the current at time k (in A, positive values ​​indicate discharging, negative values ​​indicate charging), Δt is the sampling interval (in seconds), and C_n is the charge corresponding to a 1% change in state of charge (in As). The estimated time timestamp in the initial state of charge value is used to determine the starting time of the state recursion. From this time, the state prediction value is updated progressively forward according to the sampling interval. The predicted state of charge is calculated according to the state transition equation at each sampling time, and the state covariance matrix is ​​updated synchronously according to the error propagation law. The predicted state of charge is output in time series form. Each sampling point contains the predicted state of charge value and the corresponding prediction covariance. The prediction covariance reflects the prediction uncertainty based solely on the state transition model.

[0041] The state correction and estimated residual sequence are generated based on the deviation between the predicted state of charge (SOC) and the terminal voltage signal. The predicted SOC is mapped to the predicted terminal voltage through a battery equivalent circuit model. The predicted terminal voltage equals the open-circuit voltage corresponding to the predicted SOC minus the ohmic voltage drop and polarization voltage. The open-circuit voltage is obtained by looking up the open-circuit voltage lookup table in the circuit parameters. The difference between the measured terminal voltage signal and the predicted terminal voltage is defined as the observation residual. At each sampling time, the terminal voltage signal provides a measured voltage value for comparison with the predicted value. The observation residual reflects the degree of deviation between the current predicted SOC and the actual state. The correction amount for the predicted SOC is calculated based on the observation residual and the Kalman gain. The Kalman gain is determined by the ratio of the prediction covariance to the observation noise variance. A larger prediction covariance results in a larger Kalman gain, indicating greater trust in the observed value rather than the predicted value. The state correction amount equals the Kalman gain multiplied by the observation residual. The sign of the correction amount is consistent with the sign of the observation residual; a positive observation residual corresponds to a positive correction amount, indicating that the predicted SOC needs to be adjusted upwards. The terminal voltage signal generates an observation residual at each sampling time. The observation residuals from all sampling times are arranged in chronological order to form an estimated residual sequence. The estimated residual sequence is stored in array form, where each element is the observation residual value at each sampling time. The array length increases with the number of filtering iterations. The state correction is output synchronously with the estimated residual sequence. The state correction is used to correct the predicted state of charge at the current time, while the estimated residual sequence is used for statistical characteristic analysis.

[0042] The predicted state of charge (SOC) is iteratively corrected using a state correction factor to form the real-time SOC. The state correction factor is added to the predicted SOC to obtain the corrected SOC estimate, which integrates the prediction information from the state transition model and the correction information from the terminal voltage observations. During the correction process, the predicted SOC synchronously updates the state covariance matrix. The updated covariance matrix reflects the estimation uncertainty after incorporating observation information, which is typically less than the uncertainty based solely on the prediction. The magnitude of the state correction factor is constrained to prevent estimation jumps caused by anomalous observations. The absolute value of a single correction is set to an upper limit of 5% of the SOC; values ​​exceeding this limit are truncated. Iterative correction repeats the prediction-correction loop at each sampling time. The corrected SOC predicted value serves as the starting point for the state recursion at the next time step, forming a closed-loop structure for recursive estimation. The real-time SOC is defined as the estimated SOC after iterative correction at the current sampling time, which integrates historical state information and current observation information. The statistical characteristics of the correction values ​​at each time step in the state correction sequence reflect the degree of matching between the prediction model and the observation model. A sustained deviation of the mean correction value from zero indicates a systematic bias requiring model parameter adjustments. The real-time state of charge is output as a percentage, with the estimated covariance at the current time step included as an uncertainty index.

[0043] In some embodiments, generating an estimated confidence interval based on the statistical characteristics of the estimated residual sequence includes: performing sliding window statistics on the estimated residual sequence to obtain the residual mean and residual variance; identifying the estimated bias component based on the residual mean to generate a bias correction parameter; generating an initial confidence boundary width by combining the residual variance with a preset confidence level; and dynamically fusing the bias correction parameter and the initial confidence boundary width to form an estimated confidence interval.

[0044] The residual mean and variance are obtained by performing a sliding window statistical analysis on the estimated residual sequence. A fixed-length sliding window is used for local statistical analysis of the estimated residual sequence, with the window length set to 100 sampling points corresponding to 1 second of data at a sampling frequency of 100Hz. This window length strikes a balance between statistical stability and real-time responsiveness. The sliding window moves point by point across the estimated residual sequence, recalculating the mean and variance of the residual data within the window after each sampling point movement. The residual mean is the arithmetic mean of the residual values ​​within the sliding window, reflecting the level of systematic deviation between the predicted and measured voltages. A consistently positive residual mean indicates a lower predicted voltage, a consistently negative mean indicates a higher predicted voltage, and a mean fluctuating around zero indicates a good match between the prediction model and the actual system. The residual variance is calculated based on the dispersion of the residual values ​​within the window relative to the mean, reflecting the random fluctuation amplitude of the observed residuals. The magnitude of the residual variance in the estimated residual sequence is related to the observation noise level and the degree of model mismatch. A sudden increase in variance may indicate a drastic change in battery operating conditions or a sensor malfunction. The residual mean and residual variance are output synchronously in time series form, with the series length being one window length shorter than the original estimated residual sequence.

[0045] Bias correction parameters are generated based on the identification of estimated bias components using the residual mean. The phenomenon of a persistent deviation of the residual mean from zero is defined as estimated bias, which originates from the difference between the battery model parameters and the actual parameters, or from zero-point drift of the sensor. The estimated bias component is extracted from the residual mean sequence through a low-pass filter. The filter cutoff frequency is set to 0.01Hz to filter out high-frequency random fluctuations, retaining only the slowly changing bias trend. The estimated bias component reflects the long-term systematic deviation component in the residual mean. When the amplitude of the estimated bias component in the residual mean exceeds a set threshold, a bias correction mechanism is triggered. The threshold is set to twice the standard deviation of the observation noise; biases below this threshold are considered normal fluctuations and are not corrected. The bias correction parameters are generated from the estimated bias component extracted from the residual mean after sign determination and amplitude quantization. When the residual mean is positive, the bias direction is marked positive, indicating that the estimated state of charge (SOC) needs to be adjusted upwards; when the residual mean is negative, the bias direction is marked negative, indicating that the estimated SOC needs to be adjusted downwards. The bias amplitude in the bias correction parameters is calculated based on the sensitivity relationship between voltage bias and state of charge (SOC). The conversion formula is ΔSOC_bias = ΔV_bias / k_ocv, where ΔSOC_bias is the bias amplitude (%SOC), ΔV_bias is the estimated bias component (mV), and k_ocv is the slope of the open-circuit voltage curve at the current operating point (mV / %SOC). The bias confidence level in the bias correction parameters is assessed based on the stability of the estimated bias component; the smaller the fluctuation of the estimated bias component, the higher the confidence level. The bias correction parameters integrate three elements: bias direction, bias amplitude, and bias confidence level. The bias direction and bias amplitude determine the direction and magnitude of the adjustment, while the bias confidence level constrains the intensity of the adjustment.

[0046] The initial confidence margin width is generated by combining the residual variance with a preset confidence level. The residual variance reflects the degree of random fluctuation of the estimated state of charge (SOC) value around the true value; a larger variance indicates higher uncertainty in the estimate, and the corresponding confidence interval should be wider. The preset confidence level is typically chosen as 95%, corresponding to a coverage range of approximately 1.96 standard deviations under a normal distribution. This confidence level provides good reliability assurance in engineering applications. The initial confidence margin width is calculated based on the square root of the residual variance, i.e., the residual standard deviation, using the formula W_init = z_α × σ_r × k_sens, where W_init is the initial confidence margin width in % SOC, z_α is the standard normal quantile corresponding to the confidence level (1.96 at the 95% confidence level), σ_r is the residual standard deviation in mV, and k_sens is the voltage-state-of-charge sensitivity coefficient in %SOC / mV. The residual variance exhibits different numerical levels under different operating conditions. Under high current conditions, the residual variance is typically larger, while under static or low current conditions, it is smaller. Therefore, the initial confidence margin width dynamically changes with the operating conditions. When the residual variance suddenly increases, the initial confidence margin width widens accordingly to maintain the effectiveness of the confidence level; after the residual variance returns to normal, the margin width narrows accordingly. The initial confidence margin width is output in time series form, maintaining the same time resolution and series length as the residual variance series.

[0047] The estimated confidence interval is formed by dynamically fusing the bias correction parameters and the initial confidence boundary width. The bias magnitude in the bias correction parameters adjusts the center position of the confidence interval; a positive bias shifts the entire confidence interval upwards, while a negative bias shifts it downwards, with the shift equal to the bias magnitude. The initial confidence boundary width determines half the width of the confidence interval: the upper boundary equals the center position plus half the width, and the lower boundary equals the center position minus half the width. The bias confidence level in the bias correction parameters affects the actual execution magnitude of the bias correction; at high confidence levels, the bias correction is executed at the full bias magnitude, while at low confidence levels, the bias correction magnitude is reduced proportionally to the confidence level to decrease the risk of incorrect correction. Dynamic fusion also considers the interaction between the bias correction parameters and the initial confidence boundary width; when the bias magnitude is large, the initial confidence boundary width is appropriately widened to cover any additional uncertainties that the bias correction may introduce. The estimated confidence interval is constructed centered on the current real-time state of charge (SOC). The estimated covariance of the SOC is also included in the calculation of the confidence interval width; the larger the covariance, the wider the interval. The symmetrical half-width in the initial confidence boundary width is transformed into asymmetrical upper and lower boundary widths under the influence of the bias correction parameter. With a positive bias, the lower boundary width is greater than the upper boundary width, and vice versa with a negative bias. The estimated confidence interval includes the lower boundary value, the upper boundary value, the center value, and a confidence level indicator. The typical range of the interval width is between ±2% and ±8% of the SOC.

[0048] Step S140: Perform long-term cumulative feature analysis on the estimated residual sequence to identify aging drift components, extract parameter compensation factors from the aging drift components, and use the parameter compensation factors to adaptively update the circuit parameters to form corrected circuit parameters.

[0049] In some embodiments, the step of performing long-term cumulative feature analysis on the estimated residual sequence to identify aging drift components includes: performing multi-period cumulative summation on the estimated residual sequence to obtain a cumulative residual curve; identifying monotonic drift trend segments from the cumulative residual curve to generate a drift trend identifier; extracting a drift slope parameter based on the drift trend identifier; and quantifying the degree of aging using the drift slope parameter to obtain the aging drift components.

[0050] The cumulative residual curve is obtained by multi-cycle cumulative summation of the estimated residual sequence. The residual values ​​of a single sample in the estimated residual sequence are significantly affected by measurement noise and transient operating conditions, making it difficult to identify slow aging drift trends through direct analysis. The cumulative summation method, through the integral effect, gradually amplifies weak systematic deviations to an observable level. This method has significant advantages in identifying model parameter drift caused by capacity decay and internal resistance increase during battery aging. Multi-cycle cumulative summation segments the estimated residual sequence according to charge-discharge cycle periods. The residual values ​​within each complete charge-discharge cycle are summed point-by-point to obtain the cumulative residual contribution for that cycle. The contributions for each cycle are then accumulated cycle-by-cycle. If the mean residual in the estimated residual sequence is zero, the cumulative residual curve exhibits a random walk characteristic around zero. If the mean residual is consistently positive, the cumulative residual curve shows a monotonically increasing trend; if the mean residual is consistently negative, it shows a monotonically decreasing trend. The trend of the curve directly reflects the direction of the systematic deviation between the battery equivalent circuit model parameters and the actual battery characteristics. The cumulative residual curve is plotted on the horizontal axis by the number of charge-discharge cycles and on the vertical axis by the cumulative residual value. The slope of the curve intuitively reflects the degree and direction of the systematic bias in the residuals. To generate a statistically significant cumulative residual curve, the amount of data needed to estimate the residual sequence must cover at least 10 complete charge-discharge cycles. Insufficient data reduces the reliability of the cumulative residual curve, making it difficult to distinguish between true aging drift and random fluctuations. The number of data points in the cumulative residual curve equals the number of complete charge-discharge cycles, with each data point recording the total cumulative residual value up to that cycle.

[0051] For example, the step of identifying monotonic drift trend segments from the cumulative residual curve to generate drift trend identifiers includes: performing piecewise linear fitting on the cumulative residual curve to obtain a piecewise slope sequence; identifying segments with consistent signs from the piecewise slope sequence to form a set of monotonic segments; calculating the rate of change of slope of adjacent segments based on the set of monotonic segments to identify aging acceleration features; and filtering significant drift segments through the aging acceleration features to generate drift trend identifiers.

[0052] Piecewise linear fitting is performed on the cumulative residual curve to obtain a piecewise slope sequence. The cumulative residual curve is linearly fitted piecewise according to a fixed-length window of 10 charge-discharge cycles. A 50% overlap rate is used between adjacent windows to improve the smoothness and continuity of the slope estimation. This overlap rate ensures local feature capture while avoiding abrupt changes in slope estimation between adjacent windows. Each element of the piecewise slope sequence corresponds to the linear fitting slope value of a window. The slope is calculated using the least squares method, which determines the optimal slope and intercept by minimizing the sum of squared residuals between the fitted line and the actual data points. A good fit is indicated by good linearity of the cumulative residual curve within a window, while a poor fit is indicated by significant nonlinear fluctuations, suggesting unstable drift characteristics or external disturbances within that period. The numerical range of the piecewise slope sequence is typically ±0.5 mV / cycle. A larger absolute slope value indicates more significant aging drift within that period; a positive slope indicates an increase in cumulative residuals, while a negative slope indicates a decrease. The total length of the cumulative residual curve determines the number of elements in the segmented slope sequence. After segmentation, the data from 100 charge-discharge cycles produces approximately 19 slope values. The sequence length is sufficient to support subsequent trend analysis. A sequence that is too short will make it difficult to distinguish between trends and noise, while a sequence that is too long will reduce the sensitivity to local changes.

[0053] A set of monotonic segments is formed by identifying segments with consistent signs from the piecewise slope sequence. Segments in the piecewise slope sequence where multiple consecutive elements maintain the same sign correspond to monotonic segments of the cumulative residual curve. Consistent signs indicate that the drift direction is stable and has not reversed within that period. Monotonic rising segments indicate that the model's predicted values ​​are systematically low, while monotonic falling segments indicate that the model's predicted values ​​are systematically high. The formation of the monotonic segment set involves element-by-element sign detection of the piecewise slope sequence, recording the position where the sign flips as the segment boundary. Elements between adjacent boundary points are grouped into the same monotonic segment. The identification of the sign flip position uses a zero-point detection method and considers the impact of numerical accuracy. Segments with consistent signs less than three elements in the piecewise slope sequence are considered transient fluctuations and filtered out. Only monotonic segments with a certain degree of persistence are retained in the monotonic segment set. This filtering mechanism avoids misjudging random fluctuations as valid drift trends. The monotonic segment set records the start and end positions, segment length, and mean slope of each segment. A longer segment length indicates a longer duration of the drift trend. The mean slope reflects the average intensity of the drift, and the start and end positions identify the battery usage stage where the drift occurs. If a monotonic segment in the segmented slope sequence has a length exceeding half the total sequence length, it is marked as the dominant drift segment, and its mean slope has the highest representativeness. The dominant drift segment usually corresponds to the main stage of battery aging. The number of elements in the monotonic segment set reflects the complexity of the cumulative residual curve. Fewer elements indicate a stable drift trend and a relatively simple battery aging process, while more elements indicate frequent changes in the drift direction, potentially indicating competition from multiple aging mechanisms or drastic changes in operating conditions.

[0054] Accelerated aging characteristics are identified by calculating the slope change rate of adjacent segments in a set of monotonic segments. The slope change rate is obtained by dividing the difference in the mean slope of two adjacent segments in the set of monotonic segments by the number of cycles between the two segments. The slope change rate reflects the rate of change of the drift trend over time; a positive value indicates that the drift slope is gradually increasing, while a negative value indicates that the drift slope is gradually decreasing or reversing. The slope change rate is the core criterion for accelerated aging characteristics. When the slope change rate is consistently positive and its absolute value exceeds a set threshold, it is considered accelerated aging, indicating that the battery may be entering an accelerated degradation stage. This stage usually corresponds to the accumulation of internal structural damage in the battery reaching a critical point, the rate of loss of active materials accelerates, or the solid electrolyte interface film thickens rapidly, leading to an increased rate of capacity decay and internal resistance growth. When the slope change rate of multiple consecutive pairs of adjacent segments in the set of monotonic segments remains positive, the accelerated aging characteristic is marked as significant. Continuity requires at least 3 pairs of adjacent segments to meet the acceleration condition to exclude the influence of random fluctuations. The aging acceleration characteristics are distinguished by the sign of the rate of change of the slope. Positive acceleration corresponds to accelerated capacity decay, usually accompanied by a rapid decrease in usable battery capacity. Negative acceleration corresponds to accelerated internal resistance growth, usually accompanied by an increase in the charge-discharge voltage difference and a decline in power capability. These two types reflect different dominant aging mechanisms. The mean slope of the dominant drift segment in the monotonic segment set is used as a benchmark reference in the aging acceleration characteristic analysis to assess the relative magnitude of the acceleration. If the rate of change of the slope of each adjacent segment in the monotonic segment set fluctuates below the threshold, the aging acceleration characteristic is marked as insignificant, indicating that the battery is still in the aging plateau period, the drift rate is relatively stable, and the battery performance degradation is slow and highly predictable.

[0055] Drift trend identifiers are generated by filtering significant drift segments using accelerated aging features. Monotonic segments within accelerated aging intervals marked as significant are automatically included in significant drift segments. Accelerated intervals have the highest priority because they correspond to periods of rapid battery performance degradation, requiring close monitoring and timely parameter adjustments. Monotonic segments outside accelerated intervals are filtered based on length and slope amplitude. Segments exceeding 20 cycles in length or with an absolute slope value exceeding 1.5 times the overall mean are included in significant drift segments. This dual filtering ensures that both long-duration, slow drifts and short-duration, high-intensity, rapid drifts are captured. The drift trend identifier assigns weights to each selected significant drift segment according to its slope amplitude; segments with larger absolute slope values ​​have higher weights. These weights are used to calculate a weighted average drift slope as an overall drift evaluation indicator. This weighted averaging method avoids situations where a single extreme segment dominates the overall evaluation. The acceleration type information in the aging acceleration characteristics is transmitted to the drift trend indicator. If the acceleration type is positive, the overall indicator indicates that capacity decay is dominant; if it is negative, the indicator indicates that internal resistance growth is dominant. This indicator guides the focus of subsequent parameter compensation. The overall drift assessment index in the drift trend indicator comprehensively reflects the long-term trend of the cumulative residual curve. A positive value indicates that the predicted state of charge is systematically too high, requiring downward correction of relevant parameters; a negative value indicates that the predicted state of charge is systematically too low, requiring upward correction of relevant parameters. The absolute value reflects the severity of the deviation and the urgency of correction.

[0056] The drift slope parameter is extracted based on the drift trend identifier. The overall drift assessment index in the drift trend identifier is directly used as the core value of the drift slope parameter. This value is calculated by weighted averaging of each significant segment, comprehensively reflecting the direction and amplitude of long-term drift. Compared with the slope estimation of a single period, it has stronger representativeness and anti-interference ability. The sign of the drift slope parameter is consistent with the slope sign of the dominant drift segment in the drift trend identifier. Positive values ​​correspond to a continuous upward trend in the cumulative residual, and negative values ​​correspond to a continuous downward trend in the cumulative residual. The consistency of the sign ensures the correctness of the parameter correction direction. The dispersion of the slope of each significant segment in the drift trend identifier determines the stability of the drift slope parameter. The dispersion is quantified by calculating the ratio of the standard deviation of the slope of each segment relative to the weighted average to the mean. When the dispersion is less than 20%, the stability is high, indicating that the drift direction and intensity of each significant segment are relatively consistent. When the dispersion exceeds 50%, the stability is low, indicating that the drift characteristics of each significant segment are significantly different. The stability affects the confidence assessment of subsequent aging quantification. The drift slope parameter is determined by the ratio of the total number of cycles covered by the significant segment in the drift trend indicator to the total number of cycles in the cumulative residual curve. A higher coverage rate indicates a more prevalent drift phenomenon and a more representative drift slope parameter. A coverage rate above 70% indicates that the drift characteristic persists throughout the observation period, while a coverage rate below 30% suggests that the drift characteristic is not obvious, possibly indicating that the battery is still in the early stages of aging or is being used under relatively mild conditions. If accelerated aging characteristics are present in the drift trend indicator, the drift slope parameter will be marked with an acceleration flag, indicating that the battery may be in the later stages of aging and requires more frequent parameter updates. The acceleration flag also triggers a warning about the battery's health status, reminding users to pay attention to the risk of accelerated battery performance degradation.

[0057] The aging drift component is obtained by quantifying the degree of aging through the drift slope parameter. An empirical mapping relationship exists between the core value of the drift slope parameter and the degree of battery aging. This mapping relationship is established by conducting accelerated aging experiments on similar batteries and tracking the correspondence between their residual drift characteristics and actual aging indicators. The larger the absolute value of the slope, the more severe the model mismatch caused by aging, and the greater the parameter correction required. The aging drift component converts the drift slope parameter into an aging indicator with clear physical meaning. A positive slope is mainly associated with capacity decay, and is converted into a capacity decay indicator. The physical root cause of capacity decay is the reduction of lithium-ion insertion sites in the graphite interlayer of the negative electrode and the degradation of the lattice structure of the positive electrode active material. A negative slope is mainly associated with internal resistance growth, and is converted into an internal resistance growth indicator. The physical root cause of internal resistance growth is the continuous growth and thickening of the solid electrolyte interface film and the propagation of microcracks inside the electrode. The stability and coverage of the drift slope parameter jointly determine the quantification confidence of the aging drift component. A high level of confidence is achieved when both are high, and a corresponding decrease in confidence is achieved when either is low. The aging drift component determines the dominant aging type based on the sign of the drift slope parameter and the acceleration type in the drift trend indicator. When the dominant type is capacity decay, subsequent parameter compensation focuses on capacity correction; when the dominant type is internal resistance increase, it focuses on internal resistance correction. When dual aging mechanisms coexist, both capacity and internal resistance parameters need to be corrected simultaneously to comprehensively reflect the battery's aging state. When an acceleration marker is included in the drift slope parameter, both the capacity decay indication and internal resistance increase indication of the aging drift component are amplified by an acceleration factor. The acceleration factor is determined based on the magnitude of the slope change rate to ensure that parameter compensation can keep up with the pace of aging acceleration.

[0058] Parameter compensation factors are extracted from the aging drift component. The aging drift component reflects the degree of parameter drift caused by the loss of active materials and the thickening of the solid electrolyte interface film during long-term battery use. The greater the drift, the more severe the battery aging, requiring a larger correction to the circuit parameters. There is an approximately linear correspondence between the capacity decay indication in the aging drift component and the actual capacity loss of the battery. Each unit increase in the capacity decay indication corresponds to approximately 0.5% to 1.5% capacity decay, with the specific proportional coefficient determined based on the battery type and aging stage. The parameter compensation factor extracts two core correction factors from the aging drift component: a capacity correction coefficient and an internal resistance correction coefficient. The capacity correction coefficient is calculated based on the capacity decay indication, and the internal resistance correction coefficient is calculated based on the internal resistance growth indication. When the dominant aging type in the aging drift component is capacity decay, the adjustment range of the capacity correction coefficient is greater than that of the internal resistance correction coefficient; conversely, when the dominant aging type is internal resistance growth, the adjustment range is smaller. This mechanism ensures that parameter correction focuses on the current main aging mode of the battery. The quantization confidence level in the aging drift component is passed to the parameter compensation factor. At high confidence levels, the correction coefficient is output directly as the calculated value; at low confidence levels, the correction coefficient shrinks towards 1.0 to reduce the risk of incorrect correction. The shrinkage ratio is negatively correlated with the confidence level. The typical value range for the capacity correction coefficient in the parameter compensation factor is 0.7 to 1.0; a smaller value indicates more severe capacity decay. The typical value range for the internal resistance correction coefficient is 1.0 to 1.5; a larger value indicates a more significant increase in internal resistance.

[0059] The circuit parameters are adaptively updated using a parameter compensation factor to form the corrected circuit parameters. The capacity correction coefficient in the parameter compensation factor is multiplied by the battery's nominal capacity to obtain the corrected actual usable capacity. This capacity value reflects the battery's current true energy storage capacity, taking into account the capacity decay effect caused by long-term use. The ohmic internal resistance in the circuit parameters is multiplied by the internal resistance correction coefficient in the parameter compensation factor to obtain the corrected ohmic internal resistance. An internal resistance correction coefficient of 1.2 indicates that the ohmic internal resistance has increased by 20% compared to the initial value. The internal resistance correction coefficient in the parameter compensation factor is also applied to the two polarization resistance parameters of the circuit parameters. The first and second RC network polarization resistances are both amplified and corrected by the same proportion. This processing is based on the assumption that the aging rates of ohmic internal resistance and polarization internal resistance are usually correlated. The open-circuit voltage lookup table and the two time constant parameters in the circuit parameters remain unchanged during the adaptive update. The open-circuit voltage characteristic is less affected by aging, and the change in the time constant is slower than the change in the resistance parameters. The corrected circuit parameters add a corrected capacity value field to the original structure. This field records the actual usable capacity after aging compensation and is used for calculating the remaining capacity. The overall confidence level of the corrected circuit parameters is determined by the corrected confidence level of the parameter compensation factor and the original confidence level of the circuit parameters. The overall confidence level is high when both are at a high level, and the overall confidence level is lower when either is at a low level.

[0060] Step S150: Generate the remaining capacity value based on the corrected circuit parameters and the real-time state of charge, and output the detection result based on the remaining capacity value and the estimated confidence interval.

[0061] In some embodiments, generating the remaining capacity value based on the corrected circuit parameters and the real-time state of charge includes: obtaining a current available capacity reference based on the corrected circuit parameters; identifying a capacity attenuation coefficient through the corrected circuit parameters to generate a capacity correction factor; using the capacity correction factor to attenuate and correct the current available capacity reference to form an actual available capacity; and generating the remaining capacity value by proportionally converting the real-time state of charge and the actual available capacity.

[0062] The current available capacity benchmark is obtained based on the corrected circuit parameters. The corrected circuit parameters directly contain a corrected capacity value field, which records the actual usable capacity of the battery after aging compensation calculations. This field serves as the primary data source for the current available capacity benchmark, avoiding the complex process of calculating capacity decay from scratch. After reading the corrected capacity value from the corrected circuit parameters, the current available capacity benchmark requires temperature correction. At low temperatures, the diffusion rate of lithium ions in the electrolyte decreases, leading to a temporary decrease in the battery's usable capacity. For every 10°C decrease in temperature, the usable capacity decreases by approximately 5%-10%, but this decrease can be recovered after the temperature rises. The overall confidence level of the corrected circuit parameters affects the reliability of the current available capacity benchmark. At high confidence levels, the corrected capacity value is directly used as the benchmark. At low confidence levels, a weighted average of the corrected capacity value and the nominal capacity is used to reduce the error risk from a single data source. The weighting ratio is determined based on the overall confidence level. The current usable capacity baseline typically ranges from 70% to 100% of the nominal capacity. A value below 70% indicates that the battery has entered the late stage of aging, with significant loss of active materials. Users are advised to consider replacing the battery to ensure safe use and stable range. If the corrected capacity value field in the corrected circuit parameters is abnormal or missing, the current usable capacity baseline degenerates into a conservative estimate of the nominal capacity multiplied by an empirical aging factor. The empirical aging factor is obtained by looking up a table based on the battery's cumulative years of use and the number of charge-discharge cycles.

[0063] A capacity correction factor is generated by identifying the capacity decay coefficient through the corrected circuit parameters. The ratio of the corrected ohmic internal resistance to the initial ohmic internal resistance at the time of manufacture reflects the degree of battery aging. There is a physical correlation between internal resistance growth and capacity decay; a larger internal resistance growth usually corresponds to more severe loss of active material and capacity reduction. The capacity decay coefficient is calculated from the internal resistance growth information in the corrected circuit parameters using an empirical correlation model. The model establishes a mapping relationship between the internal resistance growth rate and the capacity retention rate. The value of the capacity decay coefficient reflects the proportion of current capacity retention relative to the initial capacity. The growth of the two polarization resistors in the corrected circuit parameters is also included in the calculation of the capacity decay coefficient. The growth of polarization resistance mainly reflects aging mechanisms such as the thickening of the solid electrolyte interface film and the blockage of electrode pores. These mechanisms lead to increased polarization and a decrease in reversible capacity. The capacity correction factor is normalized based on the capacity decay coefficient to ensure that the value of the correction factor is limited to between 0 and 1. The closer the value is to 0, the greater the capacity reduction required; a value of 1 indicates that no correction is needed. The overall confidence level of the corrected circuit parameters affects the adjustment strategy of the capacity correction factor. At low confidence levels, the capacity correction factor conservatively shrinks towards 1.0 to avoid over-correction of capacity due to parameter estimation errors. The numerical relationship between the capacity decay coefficient and the capacity correction factor is that the capacity correction factor equals 1 minus the absolute value of the capacity decay coefficient; the two complement each other in describing the capacity status.

[0064] The actual usable capacity is calculated by applying a capacity correction factor to the current available capacity baseline to correct for capacity decay. Multiplying the capacity correction factor by the current available capacity baseline yields a preliminary calculated value for the actual usable capacity. This multiplication incorporates the capacity decay effect caused by aging into the usable capacity estimate, making the estimate closer to the battery's true state. The current available capacity baseline may be underestimated after correction by the capacity correction factor. To prevent an overly pessimistic estimate of the actual usable capacity from negatively impacting user experience, a lower limit protection threshold is set at 50% of the nominal capacity. Below this threshold, a battery health warning is triggered instead of further lowering the capacity estimate. The actual usable capacity also needs to consider the impact of the discharge cutoff voltage setting on the usable capacity range. A higher discharge cutoff voltage setting results in a narrower usable state-of-charge range and a smaller usable capacity. The standard cutoff voltage condition is implicitly considered in the calculation of the capacity correction factor. The temperature correction effect in the current available capacity baseline is passed to the actual usable capacity after the multiplication operation. The actual usable capacity also reflects the usable charge level under the current temperature conditions, eliminating the need for repeated temperature compensation. The actual usable capacity is output in ampere-hours (AHs), with numerical precision retained to two decimal places to meet engineering calculation requirements. The capacity correction factor plays a crucial role as a core adjustment parameter throughout the decay correction process, ensuring the accuracy of capacity estimation at different aging stages. The calculated actual usable capacity is stored in a cache for subsequent proportional conversion, avoiding redundant calculations and improving detection efficiency.

[0065] The remaining capacity value is generated by proportionally converting the real-time state of charge (SOC) to the actual available capacity. The SOC represents the battery's current charge level as a percentage of the available capacity. Dividing the percentage by 100 to convert it to a decimal and multiplying it by the actual available capacity yields the remaining capacity value. This conversion directly reflects the linear correspondence between SOC and remaining capacity. The estimation covariance of the SOC is then transferred to the uncertainty calculation of the remaining capacity value. The estimation covariance reflects the statistical uncertainty of the SOC estimation; a larger covariance indicates lower reliability of the SOC estimate and higher uncertainty in the remaining capacity. The uncertainty of the remaining capacity value is jointly determined by the uncertainty of the actual available capacity and the uncertainty of the SOC. Both are calculated using the root mean square formula for error propagation, and the combined uncertainty is typically within the range of 3%-8% of the remaining capacity value. When the SOC approaches the extreme regions of 0% or 100%, the slope characteristics of the open-circuit voltage curve cause changes in the sensitivity of the SOC estimation, potentially increasing the relative uncertainty of the remaining capacity value in these regions. After the remaining capacity value is converted proportionally, a reasonableness check is performed. The check includes whether the value is negative, whether it exceeds the actual available capacity limit, and whether there is an abnormal jump compared with the estimated value at the previous moment. If any check condition is not met, a data anomaly flag is triggered and the fault tolerance process is started.

[0066] The test results are output based on the remaining capacity value and the estimated confidence interval. The remaining capacity value, as the core data item of the test results, directly answers the question of how much charge the battery can still provide. It is a key basis for the battery management system to predict range and allocate energy, allowing users to determine whether charging is needed and the remaining driving range. The estimated confidence interval provides a reliability boundary for the remaining capacity value. The center value of the estimated confidence interval represents the expected level of the state of charge estimation, while the upper and lower boundary values ​​determine the boundaries of the confidence range. Multiplying the remaining capacity value by the proportion of the upper and lower boundaries of the estimated confidence interval yields the confidence range of the remaining capacity. The width of the confidence range reflects the degree of uncertainty in the estimation result; the narrower the width, the more reliable the estimation. The confidence level indicator records the statistical confidence degree corresponding to this confidence range. The test results integrate and encapsulate the remaining capacity value and the estimated confidence interval, while also incorporating auxiliary information such as the timestamp of the test, the current battery operating status, and parameter health assessments, forming a complete test report data structure. When the remaining capacity value is lower than a set safety threshold, the test results include a low-battery warning indicator. The threshold is typically set at 20% of the rated capacity. The warning indicator reminds users to charge in time to avoid over-discharge damage to the battery and extend its lifespan. When the width of the estimated confidence interval exceeds the set limit, the detection result is accompanied by an estimation uncertainty warning, indicating that the reliability of the current estimation result is low. Users are advised to perform static calibration to recalibrate the state of charge estimate when conditions permit. The detection results are encapsulated and output according to a standard data format, including four components: numerical domain, confidence domain, time domain, and state domain. This facilitates data parsing and function calls by upper-layer application systems, supporting various application scenarios such as vehicle instrument display and remote monitoring.

[0067] To implement the above-described method embodiments, a method for online detection of remaining capacity of a lithium battery is provided to achieve the corresponding functions and technical effects. See also... Figure 2 , Figure 2 This diagram illustrates a structural block diagram of an online lithium battery remaining capacity detection device 200 according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The online lithium battery remaining capacity detection device 200 provided in this embodiment includes: The signal fusion module 201 is used to acquire the battery's terminal voltage signal, charging and discharging current signal, and temperature signal, and to synchronously sample and fuse the terminal voltage signal and the charging and discharging current signal to form a voltage-current joint feature; The parameter identification module 202 is used to extract the polarization voltage component based on the voltage-current joint feature and the temperature signal, identify the battery equivalent circuit parameters through the polarization voltage component to obtain the circuit parameters, and use the circuit parameters to estimate the initial state of charge to obtain the initial value of the state of charge. The state filtering module 203 is used to perform state recursive filtering using the initial state of charge to form a predicted state of charge, generate a state correction amount and an estimated residual sequence based on the deviation between the predicted state of charge and the terminal voltage signal, iteratively correct the predicted state of charge using the state correction amount to form a real-time state of charge, and generate an estimated confidence interval based on the statistical characteristics of the estimated residual sequence. The aging compensation module 204 is used to perform long-term cumulative feature analysis on the estimated residual sequence to identify aging drift components, extract parameter compensation factors from the aging drift components, and use the parameter compensation factors to adaptively update the circuit parameters to form corrected circuit parameters. The capacity output module 205 is used to generate a remaining capacity value based on the corrected circuit parameters and the real-time state of charge, and to output the detection result based on the remaining capacity value and the estimated confidence interval.

[0068] The aforementioned online lithium battery remaining capacity detection device 200 can implement the online lithium battery remaining capacity detection method of the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0069] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0070] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A method for on-line detection of the residual capacity of a lithium battery, characterized in that, include: The terminal voltage signal, charge / discharge current signal, and temperature signal of the battery are acquired, and the terminal voltage signal and the charge / discharge current signal are synchronously sampled and fused to form a voltage-current joint feature; Based on the voltage-current joint characteristics and the temperature signal, the polarization voltage component is extracted. The battery equivalent circuit parameters are identified using the polarization voltage component to obtain the circuit parameters. The initial state of charge is estimated using the circuit parameters to obtain the initial state of charge value. The initial state of charge is used to perform state recursive filtering to form a predicted state of charge. A state correction amount and an estimated residual sequence are generated based on the deviation between the predicted state of charge and the terminal voltage signal. The predicted state of charge is iteratively corrected using the state correction amount to form a real-time state of charge. An estimated confidence interval is generated based on the statistical characteristics of the estimated residual sequence. Long-term cumulative feature analysis is performed on the estimated residual sequence to identify aging drift components. Parameter compensation factors are extracted from the aging drift components, and the circuit parameters are adaptively updated using the parameter compensation factors to form corrected circuit parameters. The remaining capacity value is generated based on the corrected circuit parameters and the real-time state of charge, and the detection result is output based on the remaining capacity value and the estimated confidence interval.

2. The method according to claim 1, characterized in that, The step of synchronously sampling and fusing the terminal voltage signal and the charging / discharging current signal to form a voltage-current joint feature includes: The terminal voltage signal and the charging / discharging current signal are clock-aligned to obtain a synchronization signal pair; Self-discharge characteristic parameters are extracted from the weak current fluctuations in the quiescent state identified from the synchronization signal pair. Based on the self-discharge characteristic parameters, the synchronization signal pair is segmented according to operating conditions to form segmented response characteristics; The voltage-current joint feature is constructed using the segmented response features.

3. The method of claim 1, wherein, The extraction of the polarization voltage component based on the voltage-current joint characteristics combined with the temperature signal includes: Based on the aforementioned voltage-current joint characteristics, voltage components are separated to obtain dynamic response components; The pure polarization response is obtained by separating the ohmic voltage drop from the dynamic response components. Temperature-corrected polarization values ​​are generated by performing temperature compensation on the pure polarization response using the temperature signal. The temperature-corrected polarization value is used to quantify charge-discharge asymmetry and form a polarization voltage component.

4. The method of claim 1, wherein, The step of using the circuit parameters to estimate the initial state of charge and obtain the initial state of charge value includes: Based on the circuit parameters, a mapping relationship between open-circuit voltage and state of charge is constructed; The internal resistance characteristic parameters are extracted from the circuit parameters to generate a polarization compensation factor; A rough estimate of the state of charge is obtained by performing a rough state estimation using the mapping relationship; The initial value of the state of charge is obtained by dynamically compensating the coarse estimate of the state of charge using the polarization compensation factor.

5. The method of claim 1, wherein, The step of generating the estimated confidence interval based on the statistical properties of the estimated residual sequence includes: The mean and variance of the residuals are obtained by performing sliding window statistics on the estimated residual sequence; Based on the residual mean, the bias component is identified and estimated to generate bias correction parameters; The initial confidence boundary width is generated by combining the residual variance with a preset confidence level; The estimated confidence interval is formed by dynamically fusing the bias correction parameters and the initial confidence boundary width.

6. The method of claim 1, wherein, The step of performing long-term cumulative feature analysis on the estimated residual sequence to identify aging drift components includes: The estimated residual sequence is summed over multiple periods to obtain the cumulative residual curve; A drift trend identifier is generated by identifying monotonic drift trend segments from the cumulative residual curve; Extract the drift slope parameter based on the drift trend identifier; The degree of aging is quantified by the drift slope parameter to obtain the aging drift component.

7. The method of claim 1, wherein, The process of generating the remaining capacity value based on the corrected circuit parameters and the real-time state of charge includes: The current available capacity reference is obtained based on the corrected circuit parameters; The capacity attenuation coefficient is identified by the modified circuit parameters, and a capacity correction factor is generated. The current available capacity baseline is attenuated and corrected using the capacity correction factor to form the actual available capacity; The remaining capacity value is generated by proportionally converting the real-time state of charge with the actual available capacity.

8. The method of claim 3, wherein, The step of obtaining a pure polarization response by separating the ohmic voltage drop from the dynamic response components includes: Identify current step transient response characteristics based on the dynamic response components; The ohmic voltage drop is obtained by extracting the instantaneous voltage drop using the aforementioned current step transient response characteristics; The internal resistance characteristic parameter is generated by ratio correlation between the ohmic voltage drop and the current step transient response characteristics. Pure polarization response is obtained by voltage drop separation based on the internal resistance characteristic parameters and the dynamic response components.

9. The method of claim 6, wherein, The step of identifying monotonic drift trend segments from the cumulative residual curve and generating drift trend identifiers includes: Piecewise linear fitting is performed on the cumulative residual curve to obtain a piecewise slope sequence; From the segmented slope sequence, identify segments with consistent signs to form a set of monotonic segments; Based on the set of monotonic segments, the slope change rate of adjacent segments is calculated to identify aging acceleration characteristics; The drift trend identifier is generated by filtering out significant drift segments based on the aging acceleration characteristics.

10. An on-line detection device for the residual capacity of a lithium battery, characterized in that it comprises: include: The signal fusion module is used to acquire the battery's terminal voltage signal, charge / discharge current signal, and temperature signal, and to synchronously sample and fuse the terminal voltage signal and the charge / discharge current signal to form a voltage-current joint feature; The parameter identification module is used to extract the polarization voltage component based on the voltage-current joint feature and the temperature signal, identify the battery equivalent circuit parameters through the polarization voltage component to obtain the circuit parameters, and use the circuit parameters to estimate the initial state of charge to obtain the initial value of the state of charge. The state filtering module is used to perform state recursive filtering using the initial state of charge to form a predicted state of charge, generate a state correction amount and an estimated residual sequence based on the deviation between the predicted state of charge and the terminal voltage signal, iteratively correct the predicted state of charge using the state correction amount to form a real-time state of charge, and generate an estimated confidence interval based on the statistical characteristics of the estimated residual sequence. The aging compensation module is used to perform long-term cumulative feature analysis on the estimated residual sequence to identify aging drift components, extract parameter compensation factors from the aging drift components, and use the parameter compensation factors to adaptively update the circuit parameters to form corrected circuit parameters. The capacity output module is used to generate a remaining capacity value based on the corrected circuit parameters and the real-time state of charge, and to output the detection result based on the remaining capacity value and the estimated confidence interval.

Images