Lithium ion battery state estimation method and device

Abstract

The invention relates to the technical field of battery management systems, and discloses a lithium ion battery state estimation method and device. The method comprises the following steps: S1, carrying out online parameter identification on a parameter vector of a battery equivalent circuit model by adopting a first unscented Kalman filter to obtain an updated parameter vector; s2, calculating a predicted voltage, and calculating a voltage error between the predicted voltage and the actually measured voltage; s3, selecting a target battery model used in the current period from the first battery model and the second battery model according to the voltage error; s4, estimating the state of charge of the battery by adopting a second Kalman filter; s5, estimating the health state of the battery; s6, performing aging compensation on the updated parameter vector; s7, storing the parameter vector after aging compensation as an initial parameter vector of a next sampling period; according to the method, high-precision estimation of the state of charge and the state of health of the lithium ion battery is realized by means of cooperative work of double Kalman filters, dynamic double-model arbitration switching and aging adaptive compensation.

Description

Technical Field

[0001] This invention relates to the field of battery management system technology, and more particularly to a method and apparatus for estimating the state of lithium-ion batteries. Background Technology

[0002] With the rapid development of new energy vehicles and energy storage systems, lithium-ion batteries, as core energy storage units, directly impact the safety and reliability of the system through accurate state estimation. State of charge (SOC) and state of health (SOH) are two key parameters of battery management systems. Accurate estimation of these parameters is crucial for optimizing charging and discharging strategies, extending battery life, and ensuring safe operation.

[0003] Currently, model-based dynamic estimation methods are the mainstream approach for SOC estimation. Commonly used battery equivalent circuit models include the internal resistance model (Rint model) and the first-order equivalent circuit model (ECM model). The internal resistance model is simple to calculate and has good real-time performance, but its accuracy is relatively low; the first-order equivalent circuit model can reflect the dynamic characteristics of the battery and has higher accuracy, but its computational complexity increases. Existing technologies usually use only a single model for estimation, making it difficult to achieve a good balance between real-time performance and accuracy.

[0004] Furthermore, lithium iron phosphate batteries exhibit a distinct plateau region in their charge-discharge curves. Within this region (typically between 3.2V and 3.4V), voltage changes are extremely gradual, with a voltage change of only 0.5-1mV per unit of SOC change, far less than the sensor measurement accuracy (typically 2-5mV). This causes the voltage feedback-based Kalman filter algorithm to lose observability within the plateau region, resulting in a significant decrease in SOC estimation accuracy, with errors reaching 5-10%. Existing technologies for handling the plateau region often rely on pre-calibrated fixed parameters, which cannot be dynamically adjusted according to the battery's aging state.

[0005] Furthermore, regarding battery aging, as the number of cycles increases, parameters such as the battery's internal resistance, polarization resistance, and polarization capacitance will change, and the open-circuit voltage characteristic curve will also shift. If these parameter changes are not tracked and compensated for in a timely manner, it will lead to the accumulation of state estimation errors, affecting the reliability of the battery management system. Most existing parameter identification methods use offline calibration, which cannot achieve online adaptive updates.

[0006] Furthermore, lithium-ion batteries exhibit a significant hysteresis effect, meaning that the open-circuit voltage at the same state of charge (SOC) can vary depending on the charging and discharging history. Current technologies do not adequately address this hysteresis effect and rarely consider its evolution with battery aging.

[0007] Therefore, there is an urgent need for a battery state estimation method that can dynamically balance computational complexity and estimation accuracy, adapt to the aging characteristics of the entire battery life cycle, effectively handle the estimation problem in the plateau region, and achieve online autonomous optimization of parameters. Summary of the Invention

[0008] The purpose of this invention is to provide a method and apparatus for estimating the state of charge and health of lithium-ion batteries. It achieves high-precision estimation of the state of charge and health of lithium-ion batteries through techniques such as dual Kalman filter cooperative operation, dynamic dual-model arbitration switching, and aging adaptive compensation.

[0009] To achieve this objective, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for estimating the state of a lithium-ion battery, used to estimate the state of charge and state of health of the battery, the method comprising the following steps: S1. The parameter vector of the battery equivalent circuit model is identified online using a first unscented Kalman filter. The parameter vector includes at least internal resistance, polarization resistance, polarization capacitance, and open-circuit voltage slope. The open-circuit voltage is a function of the state of charge, and the open-circuit voltage slope is the derivative of the open-circuit voltage with respect to the state of charge. A hysteresis voltage compensation value is introduced in the voltage prediction calculation of the parameter identification. The hysteresis voltage compensation value is related to the cumulative number of cycles of the battery and the current current to obtain the updated parameter vector. S2. Calculate the predicted voltage based on the battery model and state of charge selected in the previous sampling period, and calculate the voltage error between the predicted voltage and the measured voltage. S3. Select the target battery model to be used in the current cycle from the first battery model and the second battery model according to the voltage error, wherein the first battery model is an internal resistance model and the second battery model is a first-order equivalent circuit model. S4. Using a second Kalman filter, based on the updated parameter vector and the target battery model, estimate the state of charge of the battery; S5. Estimate the health status of the battery based on the battery's cumulative ampere-hour throughput and temperature weighting coefficient; S6. Based on the health status and the chemical type of the battery, select the corresponding aging compensation model to perform aging compensation on the updated parameter vector to obtain the aging-compensated parameter vector. S7. Store the aging-compensated parameter vector as the initial parameter vector for the next sampling period; In this step, parameter identification in step S1 is performed before state estimation in step S4.

[0010] Preferably, a preset initial parameter vector and initial state of charge are used during the first execution, and the parameter vector and state of charge estimated in the previous cycle are used in subsequent cycles. Step S1 uses unscented transformation to generate Sigma sampling points for parameter identification, specifically including: Generate 2n+1 Sigma sampling points based on the parameter vector and parameter covariance matrix of the previous period, where n is the dimension of the parameter vector; For each Sigma sampling point, the predicted voltage is calculated based on the battery equivalent circuit model and the hysteresis voltage compensation value; Based on the difference between the predicted voltage and the measured voltage, the parameter vector and parameter covariance matrix are updated using Kalman gain. In step S3, the voltage prediction formula for the internal resistance model is: Vpred = OCV(SOC) - R0 × I, Where Vpred is the predicted voltage, OCV is the open-circuit voltage, SOC is the state of charge, R0 is the internal resistance, and I is the current; The voltage prediction formula for the first-order equivalent circuit model is as follows: Vpred = OCV(SOC) - R0 × I - V1, Where V1 is the polarization voltage, and the polarization voltage satisfies the differential equation: dV1 / dt=-V1 / (R1×C1)+I / C1, R1 is the polarization resistance, and C1 is the polarization capacitance.

[0011] Preferably, the method for calculating the hysteresis voltage compensation value includes the following steps: First, the hysteresis coefficient is corrected based on the battery temperature: khys eff =khys×[1+kt×(T-Tref)], Then, calculate the hysteresis voltage compensation value: Vhys=khys eff ×(1+α×Ncycle / Nref)×I, Where Vhys is the hysteresis voltage compensation value, khys is the reference hysteresis coefficient, and khys eff Here, T is the corrected hysteresis coefficient, kt is the temperature coefficient, Tref is the reference temperature, α is the cycle effect coefficient, Ncycle is the cumulative number of battery cycles, Nref is the reference number of cycles, and I is the current in the current sampling period.

[0012] Preferably, step S3 specifically includes: When the voltage error is greater than the upper limit threshold, the internal resistance model is selected as the target battery model. When the voltage error is less than the lower threshold, the first-order equivalent circuit model is selected as the target battery model. The upper limit threshold value ranges from 10mV to 15mV, and the lower limit threshold value ranges from 6mV to 10mV.

[0013] Preferably, step S3 further includes anti-jitter processing during model switching, which specifically includes: Provide a switching counter, and set the initial value of the switching counter to 0; When the model switching conditions are met, the value of the switching counter is incremented by 1; when the model switching conditions are not met, the value of the switching counter is reset to 0. Model switching is performed only when the value of the switching counter reaches a preset threshold, where the preset threshold is an integer greater than or equal to 2; When switching models, if the target model is a first-order equivalent circuit model and the previous model is also a first-order equivalent circuit model, the polarization voltage of the previous model is inherited to the current model; if the previous model is an internal resistance model, the polarization voltage is initialized to 0.

[0014] Preferably, in step S6, when the battery chemistry is a lithium iron phosphate battery, the aging compensation model includes aging compensation for the open-circuit voltage slope, internal resistance, and polarization parameters. The open-circuit voltage slope is aging compensated according to the following formula: dOCV aged =dOCV new ×(1-k1×(1-SOH)), Where SOH represents the healthy state, dOCV aged dOCV is the open-circuit voltage slope after aging compensation. new denoted as the open-circuit voltage slope of the new battery, and k1 is the open-circuit voltage slope decay coefficient of the lithium iron phosphate battery. The internal resistance is compensated for aging according to the following formula: R0 aged =R0 new ×(1+k2×(1-SOH)), Among them, R0 aged R0 is the internal resistance after aging compensation. new K is the internal resistance of the new battery, and k2 is the internal resistance growth coefficient of the lithium iron phosphate battery. The polarization parameters are compensated for aging according to the following formula: R1 aged =R1 new ×(1+k3×(1-SOH)), C1 aged =C1new ×(1-k4×(1-SOH)), Among them, R1 aged R1 is the polarization resistor after aging compensation. new K is the polarization resistance of the new battery, k3 is the aging coefficient of the polarization resistance of the lithium iron phosphate battery, and C1 is the polarization resistance of the new battery. aged For the polarization capacitor after aging compensation, C1 new k is the polarization capacitance of the new battery, and k4 is the aging coefficient of the polarization capacitance of the lithium iron phosphate battery.

[0015] Specifically, when the battery is a lithium iron phosphate battery, in the state estimation process of step S4, the state estimation method for the lithium-ion battery also performs plateau region gain adjustment processing, which specifically includes: To determine whether the battery voltage is within the plateau voltage range, the plateau voltage range is determined by the lower limit voltage and the upper limit voltage, wherein: V low =V0 low -k5×(1-SOH), V high =V0 high -k6×(1-SOH), Among them, V low V0 is the lower limit voltage of the plateau region. low Here, k5 is the nominal value of the lower limit voltage in the plateau region, and k5 is the lower limit voltage correction factor. high V0 is the upper limit voltage of the plateau region. high k6 is the nominal value of the upper limit voltage of the platform area, and k6 is the upper limit voltage correction factor; When the battery voltage is detected to be within the plateau voltage range for M consecutive sampling periods, the open-circuit voltage slope is amplified according to the following formula: dOCV eff =dOCV×G, Among them, dOCV eff The open-circuit voltage slope after gain amplification; M is the number of consecutive sampling periods, which is an integer greater than or equal to 3; and G is the gain coefficient. When the battery voltage is detected to be outside the voltage range of the platform region, the open circuit voltage slope is restored to its normal value.

[0016] Preferably, when the battery chemistry is a ternary lithium battery, the aging compensation model in step S6 includes aging compensation for internal resistance and open-circuit voltage, wherein the internal resistance is aging compensated according to the following formula: R0 aged =R0 new ×[1+k7×(1-SOH)²], Wherein, k7 is the internal resistance growth coefficient of the ternary lithium battery; The aging compensation for the open-circuit voltage specifically includes: When the battery is detected to be fully charged, i.e., the state of charge reaches 100% and the resting time exceeds the preset duration, the resting voltage is recorded, and the voltage value corresponding to the 100% state of charge in the open circuit voltage curve is updated to the resting voltage.

[0017] Preferably, after performing steps S1 to S7, the state estimation method for the lithium-ion battery further performs automatic calibration, which specifically includes: The monitoring trigger conditions include: the time interval since the last calibration reaches a time threshold, or the health status decreases by more than a health status decrease threshold relative to the last calibration time; When the triggering condition is met, the parameter vector after aging compensation is batch optimized based on historical data of voltage, current and temperature within a preset time window using the least squares method or gradient descent method. Calculate the root mean square error of voltage prediction before and after optimization. When the decrease in the root mean square error exceeds the convergence threshold, store the optimized parameter vector in non-volatile memory as the initial parameter vector for subsequent startup.

[0018] Secondly, the present invention provides a state estimation device for a lithium-ion battery, used to estimate the state of charge and state of health of the battery, characterized in that it comprises: The data acquisition unit is configured to collect the battery's voltage, current, and temperature. The parameter identification unit is configured to use a first unscented Kalman filter to identify the parameter vector of the battery equivalent circuit model online. The parameter vector includes at least internal resistance, polarization resistance, polarization capacitance and open-circuit voltage slope, wherein the open-circuit voltage is a function of the state of charge and the open-circuit voltage slope is the derivative of the open-circuit voltage with respect to the state of charge. In the identification process, a hysteresis voltage compensation value related to the cumulative number of battery cycles is introduced to obtain an updated parameter vector. The model selection unit is configured to calculate the predicted voltage based on the battery model selected in the previous sampling period and the state of charge in the previous sampling period, calculate the voltage error between the predicted voltage and the measured voltage, and select a target battery model from the first battery model and the second battery model according to the voltage error, wherein the first battery model is an internal resistance model and the second battery model is a first-order equivalent circuit model. The state estimation unit is configured to use a second Kalman filter to estimate the state of charge of the battery based on the updated parameter vector and the target battery model. A health estimation unit is configured to estimate the health status of the battery based on the battery's cumulative ampere-hour throughput and a temperature weighting factor; An aging compensation unit is configured to select a corresponding aging compensation model based on the health status and the battery's chemical type to perform aging compensation on the updated parameter vector, thereby obtaining an aging-compensated parameter vector. The parameter identification unit is executed before the state estimation unit.

[0019] Compared with the prior art, the present invention has the following beneficial effects: This invention employs a dual Kalman filter collaborative architecture that prioritizes parameter identification over state estimation. The first unscented Kalman filter optimizes the battery model parameters in real time, ensuring that the second extended Kalman filter uses the latest corrected parameters for the current cycle during state estimation. This avoids the accumulation of estimation errors caused by parameter lag in traditional methods. Simultaneously, by introducing a hysteresis voltage compensation value related to the battery's cumulative cycle count during parameter identification, the invention accurately characterizes the hysteresis effect during battery charging and discharging and its evolution with aging, further improving parameter identification accuracy and state estimation reliability. On the other hand, through the dynamic dual-model arbitration mechanism, the system intelligently switches between the internal resistance model and the first-order equivalent circuit model based on the real-time calculated voltage error: when the voltage error is large, it automatically switches to the simple internal resistance model to ensure real-time performance; when the voltage error is small, it switches to the more accurate first-order equivalent circuit model to improve accuracy. This adaptive model selection strategy significantly improves the state estimation accuracy across the entire operating range while ensuring computational efficiency. Specifically, considering the plateau characteristics of lithium iron phosphate batteries, the dual-model arbitration mechanism combined with plateau gain adjustment technology in this invention can significantly improve the estimation accuracy within the plateau region. When the battery voltage is detected to be within the plateau range, the open-circuit voltage slope gain is automatically amplified by 1.5-2.0 times to enhance the Kalman filter's response to small voltage changes. Simultaneously, a model with appropriate computational complexity is dynamically selected based on the magnitude of the voltage error to ensure real-time performance. Experimental data shows that, compared to traditional single-model methods, the SOC estimation error in the plateau region of lithium iron phosphate batteries using this invention can be reduced from 5% to less than 1%, achieving an accuracy improvement of over 80%.

[0020] On the other hand, a differentiated aging compensation model for batteries with different chemical types has been established. The corresponding aging compensation model is selected according to the health status and battery chemical type to perform aging compensation on the parameter vector, and the aging-compensated parameter vector is used as the initial parameter vector for the next sampling period. This full life cycle adaptive compensation mechanism enables the battery to maintain high-precision estimation from new product to aging state, without the need for frequent return to the factory for calibration, thus reducing operation and maintenance costs. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of the state estimation method for lithium-ion batteries according to Embodiment 1 of the present invention.

[0023] Figure 2 This is a structural block diagram of the state estimation device for a lithium-ion battery according to Embodiment 2 of the present invention. Detailed Implementation

[0024] To illustrate the possible application scenarios, technical principles, implementable specific solutions, and achievable objectives and effects of this application in detail, the following description, in conjunction with the listed specific embodiments and accompanying drawings, provides a detailed explanation. The embodiments described herein are merely illustrative of the technical solutions of this application and are therefore intended to limit the scope of protection of this application.

[0025] Example 1 Please see Figure 1 The state estimation method for lithium-ion batteries in this embodiment is mainly used to estimate the state of charge and state of health of the battery. This method can estimate the state of lithium-ion batteries such as lithium iron phosphate batteries and ternary lithium batteries, and is particularly suitable for estimating the state of lithium iron phosphate batteries. The method executes once every 10ms, forming a continuous state estimation loop, as shown in this example where the predetermined time is 10ms. Of course, 15ms, 20ms, etc., can also be selected, and can be set as needed.

[0026] The state estimation method for lithium-ion batteries includes the following steps: S1. The parameter vector θk of the battery equivalent circuit model is identified online using the first unscented Kalman filter.

[0027] The parameter vector θ here k At least the parameters include internal resistance R0, polarization resistance R1, polarization capacitance C1, and open-circuit voltage slope dOCV. The open-circuit voltage OCV is a function of the state of charge (SOC), and the OCV-SOC relationship can be obtained by looking up a table or fitting a formula. The open-circuit voltage slope dOCV is the derivative of the open-circuit voltage with respect to the state of charge, i.e., dOCV = dOCV / dSOC.

[0028] When performing this method for the first time, a preset initial parameter vector θ0 and initial state of charge SOC0 are used. The initial parameter vector θ0 is generally obtained through experimental calibration. The initial state of charge SOC0 can be estimated using the open-circuit voltage method, i.e., based on the terminal voltage V after resting. rest The parameter vector θ is obtained by consulting the OCV-SOC table. It is then used in subsequent cycles to estimate the parameter vector θ from the previous cycle. k-1 and State of Charge (SOC) k-1 .

[0029] Understandably, step S1 uses an unscented transformation to generate Sigma sampling points for parameter identification. The parameter vector θk-1 is a column vector containing multiple parameter components, θ k-1 =[R0,R1,C1,dOCV]T, where R0 is the internal resistance, R1 is the polarization resistance, C1 is the polarization capacitance, and dOCV is the open-circuit voltage slope. Based on the parameter vector θ of the previous cycle... k-1 And parameter covariance matrix P θ Generate 2n+1 Sigma sampling points θ i (i=0,1,2,...,2n), where n is the dimension of the parameter vector, and in this embodiment n=4.

[0030] Each Sigma sampling point θ i It is also a column vector containing the same parameter components, θ i =[θi,R0,θi,R1,θi,C1,θi,dOCV]T. For each Sigma sampling point θ i (i=0,1,2,...,2n), calculate the corresponding predicted voltage Vpred based on the battery equivalent circuit model and the hysteresis voltage compensation value Vhys. i .

[0031] A hysteresis voltage compensation value, Vhys, is introduced into the voltage prediction calculation based on parameter identification. This hysteresis voltage compensation value is related to the battery's cumulative cycle count Ncycles and the current current I. The calculation method for the hysteresis voltage compensation value Vhys is as follows: First, the baseline hysteresis coefficient khys is corrected based on the battery temperature T to obtain the corrected hysteresis coefficient khys. eff =khys×[1+kt×(T-Tref)]. Where, khys is the reference hysteresis coefficient, obtained through experimental calibration, kt is the temperature coefficient, and Tref is the reference temperature, usually taken as 25℃.

[0032] Then, calculate the hysteresis voltage compensation value Vhys=khyseff×(1+α×Ncycle / Nref)×I, where α is the cycle influence coefficient, usually taken as 0.3, Ncycle is the cumulative number of cycles of the battery, Nref is the reference number of cycles, usually taken as 1000 times, and I is the current of the current sampling period, positive for charging and negative for discharging.

[0033] The formula for calculating the predicted voltage is Vpred. i =OCV(SOC k )-θi,R0×I-V1 k +Vhys, where θi,R0 represents the internal resistance component in the i-th Sigma sampling point θi, i.e., the first element of the vector θi, and V1 k This is the polarization voltage of the previous cycle.

[0034] Understandably, introducing the hysteresis voltage compensation value Vhys can accurately characterize the hysteresis effect during battery charging and discharging. Specifically, during charging, Vhys is positive, predicting an increase in voltage. During discharging, Vhys is negative, predicting a decrease in voltage. Furthermore, as the battery ages, the hysteresis effect gradually strengthens, and the amplitude of the hysteresis voltage compensation value also increases accordingly, reflecting the evolution of the hysteresis effect with aging.

[0035] Based on predicted voltage Vpred i The difference between the measured voltage Vmeas and the parameter vector θ is updated using the Kalman gain. k And parameter covariance matrix P θ,k After completing one parameter identification iteration, the updated parameter vector θ is obtained. k During this process, the open-circuit voltage slope (dOCV) is simultaneously updated, providing accurate model parameters for subsequent state estimation.

[0036] S2. Calculate the predicted voltage based on the battery model selected in the previous sampling period and the state of charge in the previous sampling period, and calculate the voltage error between the predicted voltage and the measured voltage.

[0037] Specifically, the battery model selected in the previous sampling period is determined by the model selection flag. old This embodiment sets a flag. old =0 indicates the internal resistance model, Flag old =1 indicates a first-order equivalent circuit model. Based on the state of charge (SOC) of the previous sampling period. k-1 Using the battery model selected in the previous sampling period, calculate the predicted voltage Vpred. old If Flag old =0, then Vpred old =OCV(SOC k-1)-R0×I, if Flag old =1, then Vpred old =OCV(SOC k-1 )-R0×I-V1 k-1 .

[0038] Then, the predicted voltage Vpred is calculated. old The voltage error ΔV between the measured voltage Vmeas and the actual voltage Vpred is ΔV=|Vpred old -Vmeas|. This allows the voltage error ΔV to reflect the prediction accuracy of the current model, providing a basis for model selection.

[0039] S3. Select the target battery model to be used in the current cycle from the first battery model and the second battery model based on the voltage error ΔV.

[0040] The first battery model here is an internal resistance model, and the voltage prediction formula is Vpred=OCV(SOC)-R0×I, where Vpred is the predicted voltage, OCV is the open circuit voltage, SOC is the state of charge, R0 is the internal resistance, and I is the current.

[0041] Understandably, the internal resistance model is simple to calculate, has good real-time performance, and is suitable for operating conditions with large voltage errors.

[0042] The second battery model is a first-order equivalent circuit model. The voltage prediction formula is Vpred=OCV(SOC)-R0×I-V1, where V1 is the polarization voltage, and the polarization voltage V1 must satisfy the differential equation dV1 / dt=-V1 / (R1×C1)+I / C1, where R1 is the polarization resistance and C1 is the polarization capacitance.

[0043] Understandably, the first-order equivalent circuit model reflects the dynamic characteristics of the battery, has high accuracy, and is suitable for operating conditions with small voltage errors.

[0044] The model selection logic in step S3 is as follows: When the voltage error ΔV is greater than the upper threshold Vth high At that time, the internal resistance model was selected as the target battery model, i.e., Flag. new =0, where the upper threshold Vth is... high The value ranges from 10mV to 15mV, and in this embodiment, it is 12mV. When the voltage error ΔV is less than the lower threshold Vth low At that time, a first-order equivalent circuit model is selected as the target battery model, i.e., Flag. new =1, where the lower threshold Vth is... low The value ranges from 6mV to 10mV, and in this embodiment, 8mV is used. When the voltage error ΔV is within the lower limit threshold Vth low and upper limit threshold Vth high In between, maintain the model selection from the previous cycle, i.e., Flag. new =Flag old .

[0045] Furthermore, to avoid instability caused by frequent model switching, step S3 also performs anti-jitter processing during model switching. This anti-jitter processing includes the following steps: Provide a toggle counter (Counter) with an initial value of 0. When the model switching condition is met, i.e., Flag new ≠ Flag old When the time comes, the value of the switching counter will be incremented by 1, Counter = Counter + 1; When the model switching condition is not met, i.e., Flag new =Flag old When this happens, the switching counter value will be reset to 0, Counter=0; Model switching is only performed when the value of the switching counter reaches a preset threshold Cth. The preset threshold Cth is an integer greater than or equal to 2; in this embodiment, the preset threshold Cth can be 3. That is, this embodiment only performs model switching when the switching condition is met for three consecutive sampling periods, avoiding frequent model switching due to accidental measurement noise and effectively preventing erroneous operations.

[0046] Furthermore, during model switching, the state inheritance problem of polarization voltage V1 also needs to be addressed. The specific steps are as follows: If the target model is a first-order equivalent circuit model and the previous model is also a first-order equivalent circuit model, then the polarization voltage of the previous model is inherited to the current model, i.e., V1. k =V1 k-1 ; If the previous model is an internal resistance model, then the polarization voltage is initialized to 0, i.e., V1. k =0.

[0047] This approach ensures the continuity of the state during model switching, avoiding estimation errors caused by sudden state changes.

[0048] Understandably, this dynamic dual-model arbitration mechanism achieves a good balance between real-time performance and accuracy. When the battery operates under high-current charging and discharging conditions and the voltage changes rapidly, the voltage error is large. The system automatically switches to the simpler internal resistance model to ensure real-time performance. Conversely, when the battery operates under low-current or quiescent conditions and the voltage changes slowly, the voltage error is smaller. The system then switches to the more accurate first-order equivalent circuit model to improve accuracy.

[0049] S4. Employ a second Kalman filter, based on the updated parameter vector θ. k Based on the target battery model, the state of charge of the battery is estimated.

[0050] The second Kalman filter here is an extended Kalman filter. This embodiment uses the extended Kalman filter algorithm, which includes two steps: prediction and correction.

[0051] For state prediction, it is based on the state of charge (SOC) of the previous cycle. k-1 Given the current I, the prior state of charge estimate SOC^k = SOC is calculated using the ampere-hour integral method. k-1 -(ηI·Δt) / Cn, where SOC^k represents the prior estimate and η is the Coulomb efficiency. Based on the polarization voltage V1 of the previous cycle... k-1 Given the current I, the prior polarization voltage estimate V1^k is calculated using the dynamic equation of the polarization voltage.

[0052] For state correction, it is based on the target battery model and the updated parameter vector θ. k Calculate the predicted voltage Vpred, and then calculate the voltage innovation Δe = Vmeas - Vpred. Update the state of charge (SOC) using the Kalman gain. k =SOC^k+K1·Δe and polarization voltage V1 k =V1^k+K2·Δe, completing one iteration of state estimation, and obtaining the posterior state-of-charge (SOC) estimate. k and posterior polarization voltage V1 k .

[0053] S5. Estimate the battery's state of health (SOH) based on the battery's cumulative ampere-hour throughput and temperature weighting coefficient.

[0054] The cumulative ampere-hour throughput Qcum here is obtained by integrating the current I over time, Qcum = Qcum prev +|I|·Δt, where Qcum prev This represents the cumulative ampere-hour throughput of the previous cycle. The temperature weighting coefficient Kt is determined based on the battery operating temperature T and the preset temperature-aging relationship. The higher the temperature, the faster the battery aging rate, and the larger the temperature weighting coefficient.

[0055] Understandably, step S5 performs a SOH estimation every hour. This is based on the cumulative ampere-hour throughput Qcum, the temperature weighting factor Kt, and the battery's rated cycle life Ncycle. rated Calculate the current health status SOH, where Ncycle ratedRated cycle life is typically 3000 to 6000 cycles for lithium iron phosphate batteries and 1000 to 2000 cycles for ternary lithium batteries. State of Health (SOH) represents the ratio of the battery's current capacity to its rated capacity, ranging from 0 to 100%. For example, SOH=100% indicates that the battery is in brand new condition, while SOH=80% is usually used as the criterion for determining whether a battery is ready for disposal.

[0056] S6. Based on the state of health (SOH) and the battery's chemical type, select the corresponding aging compensation model to update the parameter vector θ. k Aging compensation is performed to obtain the aging-compensated parameter vector θ. aged .

[0057] The battery chemistry type is obtained through the configuration parameters of the battery management system. The battery chemistry types include two types: lithium iron phosphate batteries and ternary lithium batteries.

[0058] For lithium iron phosphate batteries, the aging compensation model includes aging compensation for open-circuit voltage slope, internal resistance, and polarization parameters. The aging compensation formula for open-circuit voltage slope is dOCV. aged =dOCV new ×(1-k1×(1-SOH)), where, dOCV aged dOCV is the open-circuit voltage slope after aging compensation. new K is the open-circuit voltage slope of the new battery, and k1 is the open-circuit voltage slope decay coefficient of the lithium iron phosphate battery, which is usually taken as 0.15.

[0059] Understandably, the aging of lithium iron phosphate batteries is mainly manifested in lithium-ion loss and active material failure, which leads to a narrowing of its OCV-SOC curve, especially the plateau region. The absolute value of the open-circuit voltage slope (dOCV) in the plateau region decreases, while its absolute value remains basically unchanged. Aging compensation of the open-circuit voltage slope is used to track the change in the plateau region slope with aging.

[0060] The aging compensation formula for internal resistance is R0. aged =R0 new ×(1+k2×(1-SOH)), where R0 aged R0 is the internal resistance after aging compensation. new Let R1 be the internal resistance of the new battery, and k2 be the internal resistance growth coefficient of the lithium iron phosphate battery, ranging from 0.2 to 0.4, and 0.3 in this embodiment. The internal resistance of the lithium iron phosphate battery increases linearly with aging. The aging compensation formula for the polarization parameters is R1. aged =R1 new ×(1+k3×(1-SOH)) and C1 aged =C1 new ×(1-k4×(1-SOH)), where R1 agedR1 is the polarization resistor after aging compensation. new C1 represents the polarization resistance of the new battery, k3 is the aging coefficient of the polarization resistance of the lithium iron phosphate battery, typically taken as 0.2 to 0.4; C1 aged For the polarization capacitor after aging compensation, C1 new k is the polarization capacitance of the new battery, and k4 is the aging factor of the polarization capacitance of the lithium iron phosphate battery, which is usually taken as 0.1 to 0.3.

[0061] To address the plateau region characteristics of lithium iron phosphate batteries, a plateau region gain adjustment process is performed during the state estimation in step S4. This plateau region gain adjustment process includes the following steps: Determine if the battery voltage Vmeas is within the plateau voltage range. The plateau voltage range is defined by the lower limit voltage V. low =V0 low -k5×(1-SOH) and upper limit voltage V high =V0 high -k6×(1-SOH) is determined, where V0low is the nominal value of the lower limit voltage of the plateau region, which is usually taken as 3.20V for lithium iron phosphate batteries. V0 high This is the nominal value of the upper limit voltage of the plateau region, usually taken as 3.40V. k5 and k6 are voltage correction coefficients used to compensate for the decreasing trend of the plateau region voltage as SOH decreases.

[0062] When the battery voltage is detected to be within the plateau voltage range for M consecutive sampling periods, i.e., V low <Vmeas<V high For M consecutive periods, based on dOCV eff =dOCV aged ×G amplifies the open-circuit voltage slope, where dOCV eff The open-circuit voltage slope after gain amplification is used to replace dOCV in steps S1 and S4. aged M is the number of consecutive sampling periods, which is an integer greater than or equal to 3. In this embodiment, it is set to 5, meaning that the voltage needs to remain in the plateau region for 5 ms to trigger gain adjustment, thus avoiding misjudgment caused by instantaneous fluctuations. G is the gain coefficient, which ranges from 1.5 to 2.0. In this embodiment, it is set to 1.8.

[0063] When the battery voltage is detected to be outside the plateau voltage range, i.e., Vmeas <V low Or Vmeas>V high To restore the open-circuit voltage slope to its normal value, i.e., using dOCV. aged Instead of dOCV eff .

[0064] Understandably, by adjusting the gain in the plateau region, the open-circuit voltage slope of the lithium iron phosphate battery in the plateau region is amplified by a factor of 1.8, enhancing the Kalman filter's response to minute voltage changes. The voltage change corresponding to a unit SOC change in the plateau region of the lithium iron phosphate battery is only 0.5 to 1 mV, far less than the sensor's measurement accuracy, causing the voltage feedback-based Kalman filter algorithm to lose observability in the plateau region. After gain amplification, the equivalent voltage change is increased, approaching or exceeding the sensor's accuracy, restoring the observability of the plateau region and significantly improving the SOC estimation accuracy of the lithium iron phosphate battery in the plateau region, reducing the error by more than 80%.

[0065] For ternary lithium batteries, the aging compensation model includes aging compensation for internal resistance and open-circuit voltage. The aging compensation formula for internal resistance is R0. aged =R0 new ×[1+k7×(1-SOH)²], where k7 is the internal resistance growth coefficient of the ternary lithium battery, with a value ranging from 0.4 to 0.6, and is taken as 0.5 in this embodiment.

[0066] Understandably, (1-SOH)² in the above formula indicates that the increase in internal resistance is not linear, but rather accelerates. This aligns with the aging mechanism in ternary batteries, where initial aging is relatively slow, but once SOH decreases to a certain level, side reactions such as continuous thickening of the SEI film and deactivation of active materials intensify, leading to a sharp increase in internal resistance.

[0067] Furthermore, for ternary lithium batteries, their OCV-SOC curve exhibits a continuous, smooth, and monotonically changing pattern. The aging compensation for open-circuit voltage is performed using the following full-charge calibration method: When the battery is detected to be fully charged (i.e., the state of charge reaches 100% and the resting time exceeds the preset duration), the resting voltage V is recorded. rest Update the voltage value corresponding to 100% state of charge in the open-circuit voltage curve to the resting voltage, and complete the aging correction of the OCV curve.

[0068] Specifically, the conditions for full charging are: SOC ≥ 99.5% and resting time ≥ 2 hours. After meeting these conditions, the resting voltage V is recorded. rest Update the OCV table: OCV(100%) = V rest Then, the OCV curve is shifted or fine-tuned proportionally to adapt the entire curve to the characteristics after aging.

[0069] S7. Transfer the aging-compensated parameter vector θ aged Store as the initial parameter vector for the next sampling period.

[0070] The parameter vector θ after aging compensation aged Including R0 aged R1 aged C1aged and dOCV aged . θ aged Assigned to θ k-1 This serves as the input parameter vector for the next sampling cycle step S1. This parameter cyclic update mechanism ensures the continuity of parameters and the sustained effectiveness of aging compensation, enabling high-accuracy estimation of the battery from new to aged states.

[0071] Furthermore, after executing steps S1 to S7, the method also performs automatic calibration processing. In this embodiment, the automatic calibration is event-triggered and not executed in every cycle. This automatic calibration processing includes the following steps: The monitoring trigger conditions include two situations: the time interval since the last calibration reaches a time threshold, which is usually 24 hours; or the health status SOH decreases more than a health status decrease threshold relative to the last calibration time, which is usually 5%.

[0072] When the triggering condition is met, based on historical voltage, current, and temperature data within a preset time window, the least squares method or gradient descent method is used to process the aging-compensated parameter vector θ. aged Batch optimization is performed. Specifically, the (V, I, T) sequence from the past 20 minutes is extracted. The objective function is constructed as the root mean square error (RMSE) of voltage prediction, and the parameter vector θ is iteratively optimized using gradient descent or least squares methods. aged This minimizes RMSE.

[0073] Calculate the root mean square error (RMS) of voltage prediction before and after optimization. Ebefore and RMSE after When the root mean square error decreases by more than the convergence threshold, the optimization is considered to have converged. The convergence threshold is typically 10%. When the optimization converges, the optimized parameter vector is stored in non-volatile memory as the initial parameter vector θ0 for subsequent startups.

[0074] In this way, every time the vehicle is powered on again, it can start from the last optimized parameters instead of the factory default parameters, realizing self-learning and self-refreshing of parameters, and never needing to return to the factory for recalibration.

[0075] If the RMSE does not converge, the parameter optimization is considered to have failed, possibly due to significant parameter drift or abnormal battery performance. In this case, a parameter drift fault is reported, and the system is downgraded to using only the internal resistance model for estimation to avoid model mismatch leading to larger estimation errors.

[0076] Example 2 Please see Figure 2The lithium-ion battery state estimation device 100 of this embodiment is used to estimate the battery's state of charge and state of health. This lithium-ion battery state estimation device 100 is mainly used in battery management systems for electric vehicles, hybrid vehicles, or energy storage systems.

[0077] The lithium-ion battery state estimation device 100 in this embodiment includes a data acquisition unit 110, a parameter identification unit 120, a model selection unit 130, a state estimation unit 140, a health estimation unit 150, and an aging compensation unit 160. Each unit is implemented as a software module in the main control chip of the battery management system and completes the real-time estimation of the battery state according to a predetermined execution order.

[0078] The data acquisition unit 110 is configured to acquire the battery's voltage, current, and temperature, with a sampling period of 1ms. The data acquisition unit 110 acquires the battery terminal voltage Vmeas in real time via a voltage sensor, the charging / discharging current I via a current sensor, and the battery surface temperature T via a temperature sensor. The acquired raw data is filtered and output as a (V, I, T) triplet, which serves as the input data for subsequent units.

[0079] The parameter identification unit 120 is configured to use a first unscented Kalman filter to perform online identification of the parameter vectors of the battery equivalent circuit model. Parameter vector θ k At least the parameters include internal resistance R0, polarization resistance R1, polarization capacitance C1, and open-circuit voltage slope dOCV, where the open-circuit voltage OCV is a function of the state of charge (SOC), and the open-circuit voltage slope dOCV is the derivative of the open-circuit voltage with respect to the state of charge. The parameter identification unit 120 must be executed before the state estimation unit 140 to ensure that the state estimation uses the latest corrected parameters for the current cycle.

[0080] During the identification process, the parameter identification unit 120 introduces a hysteresis voltage compensation value Vhys related to the cumulative number of battery cycles. Specifically, the parameter identification unit 120 first corrects the reference hysteresis coefficient khys based on the battery temperature T to obtain the corrected hysteresis coefficient khys. eff =khys×[1+kt×(T-Tref)], where kt is the temperature coefficient and Tref is the reference temperature, usually taken as 25℃.

[0081] Then, the hysteresis voltage compensation value Vhys=khys is calculated based on the corrected hysteresis coefficient, the cumulative number of battery cycles Ncycle, and the current I. eff ×(1+α×Ncycle / Nref)×I, where α is the cycle influence coefficient, usually taken as 0.3, and Nref is the reference cycle number, usually taken as 1000 times.

[0082] In the voltage prediction calculation for parameter identification, the predicted voltage formula is Vpred. k =OCV(SOC k )-I k ×R0-V1 k +Vhys, by introducing a hysteresis voltage compensation value, accurately characterizes the hysteresis effect during battery charging and discharging and its evolution with aging. Vhys is greater than zero during charging and less than zero during discharging. By setting Vhys, the offset of the hysteresis effect itself is additionally compensated, which conforms to the compensation logic of the actual hysteresis effect.

[0083] The parameter identification unit 120 uses unscented transformation to generate Sigma sampling points for parameter identification. This is based on the parameter vector θ from the previous period. k-1 And parameter covariance matrix P θ Generate 2n+1 Sigma sampling points, where n is the dimension of the parameter vector.

[0084] For each Sigma sampling point, the predicted voltage is calculated based on the battery equivalent circuit model and the hysteresis voltage compensation value. Based on the difference between the predicted voltage and the measured voltage Vmeas, the parameter vector θ is updated using the Kalman gain. k And parameter covariance matrix P θ,k This completes one parameter identification iteration, resulting in the updated parameter vector.

[0085] The model selection unit 130 is configured to calculate the predicted voltage based on the battery model selected in the previous sampling period and the state of charge in the previous sampling period, calculate the voltage error ΔV between the predicted voltage and the measured voltage, and select the target battery model from the first battery model and the second battery model according to the voltage error.

[0086] The first battery model here is an internal resistance model, and the voltage prediction formula is Vpred=OCV(SOC)-R0×I; the second battery model is a first-order equivalent circuit model, and the voltage prediction formula is Vpred=OCV(SOC)-R0×I-V1, where V1 is the polarization voltage, and the polarization voltage satisfies the differential equation dV1 / dt=-V1 / (R1×C1)+I / C1.

[0087] It is understandable that when the voltage error ΔV is greater than the upper limit threshold, the model selection unit 130 selects the internal resistance model as the target battery model. When the voltage error ΔV is less than the lower threshold, the model selection unit 130 selects a first-order equivalent circuit model as the target battery model.

[0088] The upper threshold value ranges from 10mV to 15mV, and in this embodiment, it is 12mV. The lower threshold value ranges from 6mV to 10mV, and in this embodiment, it is 8mV.

[0089] This dynamic dual-model arbitration mechanism achieves a good balance between real-time performance and accuracy. When the voltage error is large, it automatically switches to the simple internal resistance model to ensure real-time performance, and when the voltage error is small, it switches to the more accurate first-order equivalent circuit model to improve accuracy.

[0090] To avoid instability caused by frequent model switching, the model selection unit 130 also includes a jitter reduction unit. This jitter reduction unit provides a switching counter, with an initial value set to 0.

[0091] When the model switching conditions are met, the switching counter value is incremented by 1; when the model switching conditions are not met, the switching counter value is reset to 0. Model switching is only performed when the switching counter value reaches a preset threshold, which is an integer greater than or equal to 2; in this embodiment, it is 3. During model switching, if the target model is a first-order equivalent circuit model and the previous model is also a first-order equivalent circuit model, the polarization voltage V1 of the previous model is inherited to the current model; if the previous model is an internal resistance model, the approximate value of the polarization voltage is estimated based on the current data within a time window before switching, using the dynamic relationship of the first-order equivalent circuit model, as the initialization value. To reduce the transient impact of switching, a transition phase can be introduced after initialization. During this phase, soft constraints or smoothing processing are applied to the polarization voltage to ensure a smooth transition to the actual dynamic. The above strategy can be adaptively adjusted in conjunction with battery operating condition information (such as current amplitude and temperature), and noise parameter adjustment can be performed at the state estimator end to ensure the continuity and stability of state estimation.

[0092] The state estimation unit 140 is configured to use a second Kalman filter, based on the updated parameter vector θ. k Based on the target battery model, the state of charge (SOC) of the battery is estimated.

[0093] The second Kalman filter employs an extended Kalman filter, based on the target battery model and the updated parameter vector θ. k SOC estimation is completed through a two-step iterative process of prediction and correction.

[0094] Specifically, the state estimation unit 140 first performs state prediction, calculating a priori SOC estimate based on the SOC of the previous cycle and the current I. Then, it performs state correction, updating the SOC using Kalman gain based on the difference between the predicted and measured voltages to obtain the posterior SOC estimate. k .

[0095] The health estimation unit 150 is configured to estimate the battery's state of health (SOH) based on the battery's cumulative ampere-hour throughput and a temperature weighting coefficient. The cumulative ampere-hour throughput is obtained by integrating the current I over time, and the temperature weighting coefficient is determined based on the battery's operating temperature T and a preset temperature-aging relationship.

[0096] The health estimation unit performs a SOH (State of Health) estimate every hour, calculating the current SOH value by comprehensively considering factors such as the cumulative number of charge-discharge cycles, current rate, and temperature history. The SOH value represents the ratio of the battery's current capacity to its rated capacity, ranging from 0 to 100%.

[0097] The aging compensation unit 160 is configured to select the corresponding aging compensation model based on the state of health (SOH) and the battery's chemical type to update the parameter vector θ. k Aging compensation is performed to obtain the aging-compensated parameter vector θ. aged .

[0098] Specifically, in this embodiment, the aging compensation unit 160 stores two aging compensation models: lithium iron phosphate battery and ternary lithium battery, so that the device can be used with commonly used lithium-ion batteries on the market.

[0099] When the battery chemistry is lithium iron phosphate, the aging compensation model includes aging compensation for open-circuit voltage slope, internal resistance, and polarization parameters. The open-circuit voltage slope is calculated according to dOCV. aged =dOCV new Aging compensation is performed using ×(1-k1×(1-SOH)), where k1 is the open-circuit voltage slope decay coefficient of the lithium iron phosphate battery, typically taken as 0.15.

[0100] Internal resistance according to R0 aged =R0 new Aging compensation is performed using ×(1+k2×(1-SOH)), where k2 is the internal resistance growth coefficient of the lithium iron phosphate battery, with a value ranging from 0.2 to 0.4.

[0101] Polarization parameters according to R1 aged =R1 new ×(1+k3×(1-SOH)) and C1 aged =C1 new Aging compensation is performed using ×(1-k4×(1-SOH)), where k3 is the aging coefficient of the polarization resistor and k4 is the aging coefficient of the polarization capacitor.

[0102] When the battery chemistry is ternary lithium battery, the aging compensation model includes aging compensation for internal resistance and open-circuit voltage. Internal resistance is calculated according to R0. aged =R0 newAging compensation is performed using ×[1+k7×(1-SOH)²], where k7 is the internal resistance growth coefficient of the ternary lithium battery, with a value ranging from 0.4 to 0.6.

[0103] This quadratic aging model reflects the accelerated increase in the internal resistance of ternary lithium batteries. Aging compensation for the open-circuit voltage is achieved through a full-charge calibration method, which includes the following steps: When the battery is detected to be fully charged (i.e., the state of charge reaches 100% and the resting time exceeds the preset duration), the resting voltage V is recorded. rest Update the voltage value corresponding to 100% state of charge in the open-circuit voltage curve to the resting voltage, and complete the aging correction of the OCV curve.

[0104] To address the plateau region characteristics of lithium iron phosphate batteries, the aging compensation unit 160 also includes a plateau region gain adjustment function. This function is executed simultaneously during the state estimation process of the state estimation unit 140. The plateau region gain adjustment first determines whether the battery voltage Vmeas is within the plateau region voltage range, which is defined by the lower limit voltage V... low =V0 low -k5×(1-SOH) and upper limit voltage V high =V0 high -k6×(1-SOH) is determined, where V0 low This is the nominal value of the lower limit voltage of the plateau region, typically taken as 3.20V. V0 high This is the nominal value of the upper limit voltage of the plateau region, usually taken as 3.40V. k5 and k6 are voltage correction factors.

[0105] When the battery voltage is detected to be within the plateau voltage range for M consecutive sampling periods, according to dOCV eff =dOCV×G amplifies the open-circuit voltage slope, where M is the number of consecutive sampling periods, which is an integer greater than or equal to 3, and is 5 in this embodiment. G is the gain coefficient, which ranges from 1.5 to 2.0, and is 1.8 in this embodiment.

[0106] When the battery voltage is detected to exceed the plateau voltage range, the open-circuit voltage slope is restored to its normal value. By adjusting the plateau gain, the observability of the plateau region is improved by more than 2 times, thereby increasing the SOC estimation accuracy of lithium iron phosphate batteries in the plateau region.

[0107] The state estimation device 100 for a lithium-ion battery in this embodiment may further include an automatic calibration unit and a parameter storage unit. The automatic calibration unit is configured to monitor trigger conditions, which include the time interval since the last calibration reaching a time threshold, or the state of health (SOH) decreasing by more than a health state decrease threshold relative to the last calibration time.

[0108] When the triggering conditions are met, based on historical voltage, current, and temperature data within a preset time window, the parameter vectors after aging compensation are batch optimized using the least squares method or gradient descent method. The root mean square error (RMSE) of voltage prediction before and after optimization is calculated. When the RMSE decreases by more than the convergence threshold, the optimization is considered converged, and the optimized parameter vectors are stored in non-volatile memory as the initial parameter vectors for subsequent startups. If the RMSE does not converge, a parameter drift fault is reported, and the system is downgraded to using only the internal resistance model for estimation.

[0109] Combination Figure 1 and Figure 2 The present invention has the following beneficial effects: This invention employs a dual Kalman filter collaborative architecture that prioritizes parameter identification over state estimation. The first unscented Kalman filter optimizes the battery model parameters in real time, ensuring that the second extended Kalman filter uses the latest corrected parameters for the current cycle during state estimation. This avoids the accumulation of estimation errors caused by parameter lag in traditional methods. Simultaneously, by introducing a hysteresis voltage compensation value related to the battery's cumulative cycle count during parameter identification, the invention accurately characterizes the hysteresis effect during battery charging and discharging and its evolution with aging, further improving parameter identification accuracy and state estimation reliability. On the other hand, through the dynamic dual-model arbitration mechanism, the system intelligently switches between the internal resistance model and the first-order equivalent circuit model based on the real-time calculated voltage error: when the voltage error is large, it automatically switches to the simple internal resistance model to ensure real-time performance; when the voltage error is small, it switches to the more accurate first-order equivalent circuit model to improve accuracy. This adaptive model selection strategy significantly improves the state estimation accuracy across the entire operating range while ensuring computational efficiency. Specifically, considering the plateau characteristics of lithium iron phosphate batteries, the dual-model arbitration mechanism combined with plateau gain adjustment technology in this invention can significantly improve the estimation accuracy within the plateau region. When the battery voltage is detected to be within the plateau range, the open-circuit voltage slope gain is automatically amplified by 1.5-2.0 times to enhance the Kalman filter's response to small voltage changes. Simultaneously, a model with appropriate computational complexity is dynamically selected based on the magnitude of the voltage error to ensure real-time performance. Experimental data shows that, compared to traditional single-model methods, the SOC estimation error in the plateau region of lithium iron phosphate batteries using this invention can be reduced from 5% to less than 1%, achieving an accuracy improvement of over 80%.

[0110] On the other hand, a differentiated aging compensation model for batteries with different chemical types has been established. The corresponding aging compensation model is selected according to the health status and battery chemical type to perform aging compensation on the parameter vector, and the aging-compensated parameter vector is used as the initial parameter vector for the next sampling period. This full life cycle adaptive compensation mechanism enables the battery to maintain high-precision estimation from new product to aging state, without the need for frequent return to the factory for calibration, thus reducing operation and maintenance costs.

[0111] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for estimating the state of a lithium-ion battery, used to estimate the battery's state of charge and state of health, characterized in that, The state estimation method for lithium-ion batteries includes the following steps: S1. The parameter vector of the battery equivalent circuit model is identified online using a first unscented Kalman filter. The parameter vector includes at least internal resistance, polarization resistance, polarization capacitance, and open-circuit voltage slope. The open-circuit voltage is a function of the state of charge, and the open-circuit voltage slope is the derivative of the open-circuit voltage with respect to the state of charge. A hysteresis voltage compensation value is introduced in the voltage prediction calculation of the parameter identification. The hysteresis voltage compensation value is related to the cumulative number of cycles of the battery and the current current to obtain the updated parameter vector. S2. Calculate the predicted voltage based on the battery model and state of charge selected in the previous sampling period, and calculate the voltage error between the predicted voltage and the measured voltage. S3. Select the target battery model to be used in the current cycle from the first battery model and the second battery model according to the voltage error, wherein the first battery model is an internal resistance model and the second battery model is a first-order equivalent circuit model. S4. Using a second Kalman filter, based on the updated parameter vector and the target battery model, estimate the state of charge of the battery; S5. Estimate the health status of the battery based on the battery's cumulative ampere-hour throughput and temperature weighting coefficient; S6. Based on the health status and the chemical type of the battery, select the corresponding aging compensation model to perform aging compensation on the updated parameter vector to obtain the aging-compensated parameter vector. S7. Store the aging-compensated parameter vector as the initial parameter vector for the next sampling period; In this step, parameter identification in step S1 is performed before state estimation in step S4.

2. The state estimation method for lithium-ion batteries as described in claim 1, characterized in that, The initial parameter vector and initial state of charge are used during the first execution, and the parameter vector and state of charge estimated in the previous cycle are used in subsequent cycles. Step S1 uses unscented transformation to generate Sigma sampling points for parameter identification, specifically including: Generate 2n+1 Sigma sampling points based on the parameter vector and parameter covariance matrix of the previous period, where n is the dimension of the parameter vector; For each Sigma sampling point, the predicted voltage is calculated based on the battery equivalent circuit model and the hysteresis voltage compensation value; Based on the difference between the predicted voltage and the measured voltage, the parameter vector and parameter covariance matrix are updated using Kalman gain. In step S3, the voltage prediction formula for the internal resistance model is: Vpred = OCV(SOC) - R0 × I, Where Vpred is the predicted voltage, OCV is the open-circuit voltage, SOC is the state of charge, R0 is the internal resistance, and I is the current; The voltage prediction formula for the first-order equivalent circuit model is as follows: Vpred = OCV(SOC) - R0 × I - V1, Where V1 is the polarization voltage, and the polarization voltage satisfies the differential equation: dV1 / dt=-V1 / (R1×C1)+I / C1, R1 is the polarization resistance, and C1 is the polarization capacitance.

3. The state estimation method for lithium-ion batteries as described in claim 1, characterized in that, The method for calculating the hysteresis voltage compensation value includes the following steps: First, the hysteresis coefficient is corrected based on the battery temperature: khys eff =khys×[1+kt×(T-Tref)], Then, calculate the hysteresis voltage compensation value: Vhys=khys eff ×(1+α×Ncycle / Nref)×I, Where Vhys is the hysteresis voltage compensation value, khys is the reference hysteresis coefficient, and khys eff Here, T is the corrected hysteresis coefficient, kt is the temperature coefficient, Tref is the reference temperature, α is the cycle effect coefficient, Ncycle is the cumulative number of battery cycles, Nref is the reference number of cycles, and I is the current in the current sampling period.

4. The state estimation method for lithium-ion batteries as described in claim 2, characterized in that, Step S3 specifically includes: When the voltage error is greater than the upper limit threshold, the internal resistance model is selected as the target battery model. When the voltage error is less than the lower threshold, the first-order equivalent circuit model is selected as the target battery model. The upper limit threshold value ranges from 10mV to 15mV, and the lower limit threshold value ranges from 6mV to 10mV.

5. The state estimation method for lithium-ion batteries as described in claim 4, characterized in that, Step S3 further includes anti-jitter processing during model switching, which specifically includes: Provide a switching counter, and set the initial value of the switching counter to 0; When the model switching conditions are met, the value of the switching counter is incremented by 1; when the model switching conditions are not met, the value of the switching counter is reset to 0. Model switching is performed only when the value of the switching counter reaches a preset threshold, where the preset threshold is an integer greater than or equal to 2; When switching models, if the target model is a first-order equivalent circuit model and the previous model is also a first-order equivalent circuit model, the polarization voltage of the previous model is inherited to the current model; if the previous model is an internal resistance model, the polarization voltage is initialized to 0.

6. The state estimation method for lithium-ion batteries as described in claim 2, characterized in that, In step S6, when the battery chemistry is lithium iron phosphate, the aging compensation model includes aging compensation for open-circuit voltage slope, internal resistance, and polarization parameters. The open-circuit voltage slope is aging compensated according to the following formula: dOCV aged =dOCV new ×(1-k1×(1-SOH)), Where SOH represents the healthy state, dOCV aged dOCV is the open-circuit voltage slope after aging compensation. new denoted as the open-circuit voltage slope of the new battery, and k1 is the open-circuit voltage slope decay coefficient of the lithium iron phosphate battery. The internal resistance is compensated for aging according to the following formula: R0 aged =R0 new ×(1+k2×(1-SOH)), Among them, R0 aged R0 is the internal resistance after aging compensation. new K is the internal resistance of the new battery, and k2 is the internal resistance growth coefficient of the lithium iron phosphate battery. The polarization parameters are compensated for aging according to the following formula: R1 aged =R1 new ×(1+k3×(1-SOH)), C1 aged =C1 new ×(1-k4×(1-SOH)), Among them, R1 aged R1 is the polarization resistor after aging compensation. new K is the polarization resistance of the new battery, k3 is the aging coefficient of the polarization resistance of the lithium iron phosphate battery, and C1 is the polarization resistance of the new battery. aged For the polarization capacitor after aging compensation, C1 new k is the polarization capacitance of the new battery, and k4 is the aging coefficient of the polarization capacitance of the lithium iron phosphate battery.

7. The state estimation method for lithium-ion batteries as described in claim 6, characterized in that, When the battery is a lithium iron phosphate battery, during the state estimation process in step S4, the state estimation method for lithium-ion batteries also performs plateau region gain adjustment processing, which specifically includes: To determine whether the battery voltage is within the plateau voltage range, the plateau voltage range is determined by the lower limit voltage and the upper limit voltage, wherein: V low =V0 low -k5×(1-SOH), V high =V0 high -k6×(1-SOH), Among them, V low V0 is the lower limit voltage of the plateau region. low Here, k5 is the nominal value of the lower limit voltage in the plateau region, and k5 is the lower limit voltage correction factor. high V0 is the upper limit voltage of the plateau region. high k6 is the nominal value of the upper limit voltage of the platform area, and k6 is the upper limit voltage correction factor; When the battery voltage is detected to be within the plateau voltage range for M consecutive sampling periods, the open-circuit voltage slope is amplified according to the following formula: dOCV eff =dOCV×G, Among them, dOCV eff The open-circuit voltage slope after gain amplification; M is the number of consecutive sampling periods, which is an integer greater than or equal to 3; and G is the gain coefficient. When the battery voltage is detected to be outside the voltage range of the platform region, the open circuit voltage slope is restored to its normal value.

8. The state estimation method for lithium-ion batteries as described in claim 2, characterized in that, When the battery's chemical type is a ternary lithium battery, the aging compensation model in step S6 includes aging compensation for internal resistance and open-circuit voltage. The internal resistance is aging compensated according to the following formula: R0 aged =R0 new ×[1+k7×(1-SOH)²], Wherein, k7 is the internal resistance growth coefficient of the ternary lithium battery; The aging compensation for the open-circuit voltage specifically includes: When the battery is detected to be fully charged, i.e., the state of charge reaches 100% and the resting time exceeds the preset duration, the resting voltage is recorded, and the voltage value corresponding to the 100% state of charge in the open circuit voltage curve is updated to the resting voltage.

9. The state estimation method for lithium-ion batteries as described in claim 6, characterized in that, After performing steps S1 to S7, the state estimation method for the lithium-ion battery also undergoes automatic calibration, which specifically includes: The monitoring trigger conditions include: the time interval since the last calibration reaches a time threshold, or the health status decreases by more than a health status decrease threshold relative to the last calibration time; When the triggering condition is met, the parameter vector after aging compensation is batch optimized based on historical data of voltage, current and temperature within a preset time window using the least squares method or gradient descent method. Calculate the root mean square error of voltage prediction before and after optimization. When the decrease in the root mean square error exceeds the convergence threshold, store the optimized parameter vector in non-volatile memory as the initial parameter vector for subsequent startup.

10. A state estimation device for a lithium-ion battery, used to estimate the state of charge and state of health of the battery, characterized in that, include: The data acquisition unit is configured to collect the battery's voltage, current, and temperature. The parameter identification unit is configured to use a first unscented Kalman filter to identify the parameter vector of the battery equivalent circuit model online. The parameter vector includes at least internal resistance, polarization resistance, polarization capacitance and open-circuit voltage slope, wherein the open-circuit voltage is a function of the state of charge and the open-circuit voltage slope is the derivative of the open-circuit voltage with respect to the state of charge. In the identification process, a hysteresis voltage compensation value related to the cumulative number of battery cycles is introduced to obtain an updated parameter vector. The model selection unit is configured to calculate the predicted voltage based on the battery model selected in the previous sampling period and the state of charge in the previous sampling period, calculate the voltage error between the predicted voltage and the measured voltage, and select a target battery model from the first battery model and the second battery model according to the voltage error, wherein the first battery model is an internal resistance model and the second battery model is a first-order equivalent circuit model. The state estimation unit is configured to use a second Kalman filter to estimate the state of charge of the battery based on the updated parameter vector and the target battery model. A health estimation unit is configured to estimate the health status of the battery based on the battery's cumulative ampere-hour throughput and a temperature weighting factor; An aging compensation unit is configured to select a corresponding aging compensation model based on the health status and the battery's chemical type to perform aging compensation on the updated parameter vector, thereby obtaining an aging-compensated parameter vector. The parameter identification unit is executed before the state estimation unit.

Images