A battery testing method and system based on adaptive state joint estimation

Abstract

This invention relates to the field of battery management system technology, specifically to a battery testing method and system based on adaptive joint state estimation. The method includes: establishing a battery model; updating the model using the multi-innovation least squares method; iterating particles using an unscented Kalman filter algorithm based on the real-time voltage to obtain high-likelihood particles matching the real-time voltage as forward filter estimates; and performing reverse correction on the forward filter estimates to obtain a smoothed correction result; and predicting and outputting the current SOC and SOH values ​​based on the smoothed correction result. To address the problem of inaccurate state-of-charge estimation, the method introduces least-innovation squares to update the battery model parameters, accelerating parameter convergence and achieving accurate capture of model parameters. Simultaneously, by iteratively optimizing the proposal distribution function using the unscented Kalman filter (IUKF) and combining it with an RTS smoothing structure for posterior estimation, the accuracy of joint SOC and SOH estimation under fault interference environments is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of battery management system technology, and specifically to a battery testing method and system based on adaptive state joint estimation. Background Technology

[0002] Battery state-of-charge (SOC) estimation and battery state-of-health (SOH) estimation are two crucial functions in a battery management system (BMS). SOC estimation involves measuring the battery voltage within a pre-calibrated voltage range and converting it into a percentage SOC value to facilitate charge and discharge control in relevant programs. SOH estimation, on the other hand, assesses the battery's capacity degradation by evaluating factors such as internal resistance, polarization characteristics, and cycle life.

[0003] Current battery management system (BMS) testing and state estimation technologies suffer from several shortcomings, which are the core technical problems that this invention aims to solve:

[0004] Distortions in Pure Software Simulation and the Irreproducibility of Physical Environments: Existing BMS testing heavily relies on pure software simulation environments such as Simulink. While this method is inexpensive, it cannot simulate the complex characteristics of real physical connections. Pure software simulation cannot reproduce the true noise characteristics of BMS sampling circuits under electromagnetic interference (EMI), nor can it simulate the dynamic changes in contact resistance caused by wiring harness aging (i.e., "micro-open circuits" or "intermittent contacts"). In actual operating conditions, battery pack failures are often accompanied by random fluctuations in contact impedance; this non-Gaussian noise is a major cause of failure for traditional Kalman filter (KF) algorithms.

[0005] Convergence limitations of traditional parameter identification methods: In battery model parameter identification, the widely used recursive least squares (RLS) method suffers from severe data saturation. As the amount of sampled data increases, older data gradually overwhelms the features of new data, making the algorithm insensitive to parameter changes caused by battery aging (such as increased internal resistance), resulting in low identification accuracy and slow convergence. When the test platform simulates rapid load changes through relays, RLS often fails to track the dynamic response inside the battery in a timely manner, leading to a significant deviation between the model estimates and actual observations.

[0006] The "Particle Impoverishment" Particle Filtering Algorithm: For state estimation of nonlinear, non-Gaussian systems (such as lithium battery SOC estimation), the particle filter (PF) algorithm theoretically outperforms the extended Kalman filter (EKF). However, in practical applications, the traditional PF algorithm faces a severe "particle impoverishment" problem. As the number of iterations increases, high-weight particles are repeatedly copied, while low-weight particles are eliminated, leading to a loss of diversity in the particle set, eventually degenerating into a single particle. This fails to accurately cover the probability distribution of the battery state, especially when high-frequency fault interference is injected into the test platform, causing the traditional PF algorithm to easily diverge, resulting in the complete failure of SOC and SOH estimations.

[0007] These problems seriously affect the accuracy of BMS automated testing systems and pose risks to users' product usage. They are issues that all BMS manufacturers need to address. Summary of the Invention

[0008] To address the aforementioned problems in existing technologies, a battery testing method and system based on adaptive state joint estimation is provided.

[0009] The specific technical solution is as follows: A battery testing method based on adaptive state joint estimation includes: Step S1: Collecting real-time voltage and real-time current of the battery to be evaluated, and establishing a battery model corresponding to the resistive and capacitive characteristics of the battery to be evaluated based on the real-time voltage and real-time current; Step S2: Applying the multiple innovation least squares method to the battery model to predict the predicted parameters of the battery model, and updating the battery model based on the predicted parameters; Step S3: Extracting particles based on prior distribution, and iterating the particles using an unscented Kalman filter algorithm according to the real-time voltage to obtain high-likelihood particles matching the real-time voltage as forward filtering estimates; the particles represent different combinations of SOC and SOH values; Step S4: Performing reverse correction on the forward filtering estimates to obtain a smoothing correction result; Step S5: Predicting and outputting the current SOC and current SOH values ​​based on the smoothing correction result.

[0010] On the other hand, the battery model in step S1 is based on a second-order RC model, including:

[0011] ;

[0012] In the formula, The predicted terminal voltage of the battery to be evaluated. The open-circuit voltage corresponding to the current SOC value. The ohmic internal resistance of the battery to be evaluated is given. The polarization resistor of the first RC network. The polarization capacitor of the first RC network. The polarization voltage of the first RC network is denoted as . The polarization resistor of the second RC network, This is the polarization capacitance of the second RC network. denoted as the polarization voltage of the second RC network.

[0013] On the other hand, step S2 includes: step S21: discretizing the battery model to obtain a discretized model; step S22: calculating the gain matrix of the discretized model; step S23: updating the battery model according to the gain matrix to obtain parameter estimates as the prediction parameters; step S24: updating the covariance matrix of the gain matrix to converge the gain matrix at the next update.

[0014] On the other hand, step S3 includes: step S31: extracting the particles from the prior distribution and generating a proposed distribution function for particle filtering based on iterative unscented Kalman filtering; step S32: iterating the proposed distribution function based on the real-time voltage to update the mean and covariance of the particles and obtain the high likelihood particles.

[0015] On the other hand, step S4 includes: step S41: calculating a smoothing gain value for the forward filter estimate; step S42: performing a reverse recursion on the forward filter estimate based on the smoothing gain value to correct fluctuations and obtain the smoothing correction result.

[0016] On the other hand, step S5 includes: step S51: predicting the current SOC value based on the smoothing correction result; step S52: generating the current SOH value based on the current SOC value and the ohmic internal resistance in the battery model.

[0017] A battery testing system for implementing the above-described battery testing method;

[0018] The battery testing system includes: a host computer connected to the battery management system of the battery under evaluation, which measures the real-time current and voltage of the battery under evaluation; a battery cell simulator controlled by the host computer, which simulates the battery under evaluation based on the real-time current and voltage; and a fault injection circuit connected to both the host computer and the battery cell simulator. The fault injection circuit is controlled by the host computer and injects faults into the battery cell simulator. The host computer predicts the current SOC and current SOH values ​​of the battery under evaluation based on the response of the battery cell simulator to the injected fault signal under simulated conditions.

[0019] On the other hand, it also includes an insulation detection circuit; the insulation detection circuit is disposed between the input and output terminals of the battery to be evaluated and is controlled by the host computer to test the insulation of the battery to be evaluated.

[0020] On the other hand, the fault injection circuit is implemented based on a high-speed switching array.

[0021] The above technical solution has the following advantages or beneficial effects: Addressing the problem of inaccurate state of charge estimation in existing technologies, the least innovation squares method is introduced to update the battery model parameters, thereby accelerating parameter convergence and achieving accurate capture of model parameters. Simultaneously, by iteratively optimizing the proposal distribution function using the unscented Kalman spectroscopy (IUKF) and combining it with an RTS smoothing structure for posterior estimation, the joint estimation accuracy of SOC and SOH under fault disturbance environments is significantly improved. Attached Figure Description

[0022] Embodiments of the invention will be described more fully with reference to the accompanying drawings. However, the drawings are for illustration and explanation only and do not constitute a limitation on the scope of the invention.

[0023] Figure 1 This is an overall schematic diagram of an embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram of step S2 in an embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram showing the estimation results of the innovation vector at different lengths;

[0026] Figure 4 This is a schematic diagram of step S3 in an embodiment of the present invention;

[0027] Figure 5 This is a schematic diagram of step S4 in an embodiment of the present invention;

[0028] Figure 6 This is a schematic diagram of step S4 in an embodiment of the present invention;

[0029] Figure 7a This is a graph comparing the estimated and actual SOC values ​​generated during various algorithm tests under the influence of pulsed discharge.

[0030] Figure 7b This is a magnified view of the SOC estimation and actual values ​​compared to the curves generated during the testing of various algorithms under the influence of pulse discharge, within the 0-500 second range.

[0031] Figure 7c This is a magnified view of the SOC estimation and actual values ​​compared to the curves generated during the testing of various algorithms under the influence of pulse discharge, in the 3700-4200 second interval.

[0032] Figure 8 This is a graph showing the absolute error of SOC generated during the testing of various algorithms under the influence of pulse discharge.

[0033] Figure 9 This is a schematic diagram of the test system in an embodiment of the present invention;

[0034] Figure 10 This is a schematic diagram of the insulation fault test of the test system in an embodiment of the present invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0037] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0038] This invention includes: a battery testing method based on adaptive state joint estimation, such as... Figure 1 As shown, the process includes: Step S1: Acquiring real-time voltage and current of the battery to be evaluated, and establishing a battery model corresponding to the resistance and capacitance characteristics of the battery to be evaluated based on the real-time voltage and current; Step S2: Applying the multiple innovation least squares method to the battery model to predict the predicted parameters of the battery model, and updating the battery model based on the predicted parameters; Step S3: Extracting particles based on the prior distribution, and iterating the particles using the unscented Kalman filter algorithm according to the real-time voltage to obtain high-likelihood particles matching the real-time voltage as forward filtering estimates; the particles represent different combinations of SOC and SOH values; Step S4: Performing reverse correction on the forward filtering estimates to obtain a smoothing correction result; Step S5: Predicting and outputting the current SOC and current SOH values ​​based on the smoothing correction result.

[0039] Specifically, to address the problem of inaccurate state of charge (SOC) estimation in existing technologies, the least innovation squares method is introduced to update the parameters of the battery model, thereby accelerating the convergence of the battery model parameters and achieving accurate capture of the model parameters. Simultaneously, by iteratively optimizing the proposal distribution function using the unscented Kalman Array (IUKF) and combining it with the RTS smoothing structure for posterior estimation, the joint estimation accuracy of SOC and SOH under fault disturbance environments is significantly improved.

[0040] Specifically, the above-mentioned technical solutions are mainly configured as software embodiments in computer devices, such as battery management systems for batteries and host controllers for energy storage systems, to establish equivalent battery models and estimate parameters for the battery to be evaluated, until finally the current SOC value and current SOH value of the battery to be evaluated are estimated and output.

[0041] The battery under evaluation is measured using an online sampling circuit to obtain real-time voltage and current. However, considering that various faults may occur in the sampling circuit during practical applications, such as loose sampling cable plugs, power rail voltage fluctuations, external circuit failures, etc., the measured real-time voltage and current will introduce a significant amount of noise, thus masking the true signal. Directly calculating the SOC and SOH values ​​from tables based on these real-time voltage and current readings will introduce substantial errors, especially in online applications. Since SOC and SOH values ​​are typically cumulative, sudden changes in these values ​​will not only confuse users but also introduce significant cumulative errors over long-term operation.

[0042] To address this, the proposed solution employs a two-stage estimation process:

[0043] In the first stage, a battery model is first established for the battery to be estimated to simulate the battery's physical parameters, including ohmic internal resistance, polarization voltage, polarization capacitance, etc. Then, the parameters of the battery model are updated based on the multiple innovation least squares (MILS) method, and the measured real-time voltage is used as a reference to make the parameters of the battery model approach the real physical parameters of the battery to be estimated.

[0044] For the equivalent circuit model, the algorithm utilizes the "Multi-Innovation" theory, meaning that in each iteration, it uses not only the observation data at the current moment but also the innovation vectors from the past p moments to correct the parameter estimates. Compared to the traditional RLS algorithm, MILS significantly improves the identification accuracy and convergence speed of the ohmic internal resistance R_0, polarization resistance R_p, and polarization capacitance C_p by extending the innovation length (p=5,10,15). In simulation testing, it effectively overcomes the transient interference introduced by the relay operation of the test platform, thus achieving better suppression of noise caused by external circuit faults during actual measurements.

[0045] In the second stage, the system transitions to a joint estimation process of SOC / SOH for the equivalent model. Specifically, based on parameter identification, the system employs an Iterative Unscented Kalman Particle Filter (RTS-IUPF) algorithm with an RTS smoothing structure for state estimation. The Iterative Unscented Kalman Filter (IUKF) generates the proposal distribution function for the particle filter. IUKF uses the latest observations to iteratively update the particles multiple times, moving them towards the high likelihood region, thus addressing the "particle scarcity" problem of traditional PF. After forward filtering, the Rauch-Tung-Striebel (RTS) backward smoothing algorithm is introduced. Future observation information is used to correct past state estimates, further reducing the estimation error covariance. Finally, a dual-timescale or joint state vector approach is used to simultaneously estimate the rapidly changing SOC (State of Charge) and the slowly changing SOH (State of Health).

[0046] Experiments show that, under pulsed discharge and complex operating conditions, the absolute error of SOC estimation does not exceed 1.911%, and the absolute error of SOH estimation is less than 4%.

[0047] In one embodiment, the battery model in step S1 is based on a second-order RC model, including:

[0048] ;

[0049] In the formula, The predicted terminal voltage of the battery to be evaluated. The open-circuit voltage corresponding to the current SOC value. The ohmic internal resistance of the battery to be evaluated. The polarization resistor of the first RC network. The polarization capacitor of the first RC network. The polarization voltage of the first RC network. The polarization resistor of the second RC network, This is the polarization capacitor of the second RC network. This is the polarization voltage of the second RC network.

[0050] In one embodiment, such as Figure 2 As shown, step S2 includes: step S21: discretizing the battery model to obtain a discretized model; step S22: calculating the gain matrix of the discretized model; step S23: updating the battery model based on the gain matrix to obtain parameter estimates as prediction parameters; step S24: updating the covariance matrix of the gain matrix to converge the gain matrix for the next update.

[0051] Specifically, to update the parameters of the battery model, this embodiment first establishes the aforementioned second-order RC model to be equivalent to the battery under evaluation, and then searches for the model parameters. Specifically, for digital control, the aforementioned continuous equation needs to be discretized into an ARX (autoregressive ergodic) model format using bilinear transformation or the finite difference method:

[0052] ;

[0053] In the formula, This is the system output, corresponding to the difference between the terminal voltage and the open-circuit voltage. The parameter vector to be identified (including the combination coefficients of resistors and capacitors, which is achieved by splicing the above-mentioned ohmic internal resistance, polarization internal resistance of the first RC network, polarization capacitor, polarization internal resistance of the second RC network, and polarization capacitor into a multi-dimensional vector). It is a data vector.

[0054] in, The complete mathematical structure is as follows:

[0055] ;

[0056] In the formula, This is the current measurement value obtained at the moment. These are historical voltage measurements from two previous cycles. These are the historical current measurements from the first two cycles.

[0057] Historical voltage measurements describe the "inertia" or memory effect of the battery polarization voltage. The current terminal voltage of the battery depends not only on the current current but also significantly on the voltage state over the past one or two sampling periods. Historical current measurements describe the driving force of external load excitation on the system. The instantaneous ohmic voltage drop is represented by the driving force, while the historical current term drives the charge accumulation in the polarization loop through the integral effect. This combination allows for the characterization of the voltage and current in the current state.

[0058] After discretizing the model, a length of [length missing] is used. New information vector Perform a MILS-based recursive update algorithm:

[0059] .

[0060] The specific process of recursive update includes:

[0061] Calculate the gain matrix based on the covariance matrix of the previous prediction. :

[0062]

[0063] in For the past An information matrix composed of data vectors at each time point This is the covariance matrix from the previous prediction.

[0064] Based on the gain matrix The amount of update to the parameter vector during the current iteration can be determined, thereby updating the parameter estimates. :

[0065] .

[0066] In the formula, These are the parameter estimates obtained from this round of forecasting. These are the prediction parameters obtained from the previous round of prediction. Here is the gain matrix. An information vector of a predetermined length.

[0067] The matrix of predicted parameters at this point corresponds to the ohmic internal resistance, polarization internal resistance of the first RC network, polarization capacitance, polarization internal resistance of the second RC network, polarization capacitance, etc. in the aforementioned second-order RC model. The updated model can be used for the estimation process of SOC and SOH.

[0068] Finally, the covariance matrix also needs to be updated. :

[0069]

[0070] In the formula, This is the covariance matrix for this round of predictions. This is the covariance matrix from the previous prediction. Here is the gain matrix. For the past An information matrix composed of data vectors at each time point.

[0071] This step is used to ensure that the covariance matrix converges over multiple iterations, thereby controlling the "adjustment amplitude" of the gain matrix. Since the gain matrix is ​​calculated based on the covariance matrix in each round of calculation, assuming that the model parameters of the system become closer to the actual values ​​in multiple prediction rounds, the amplitude of the gain matrix can be adjusted through the converged covariance matrix to avoid parameter oscillations near the steady-state point.

[0072] like Figure 3 As shown, under HPPC conditions, when the innovation length p=15, the average absolute error of the terminal voltage prediction of the MILS algorithm is only 0.02197V, which is significantly better than the 0.03892V of the traditional RLS algorithm, proving the robustness of the method in the noisy environment of the hardware test platform.

[0073] In one embodiment, such as Figure 4 As shown, step S3 includes: step S31: extracting particles from the prior distribution and generating a proposed distribution function for particle filtering based on iterative unscented Kalman filtering; step S32: iterating the proposed distribution function based on real-time voltage to update the mean and covariance of the particles and obtain high-likelihood particles.

[0074] Specifically, in order to achieve a better forward filtering process, in this embodiment, a random particle distribution is first constructed, with each particle representing a combination of SOC and SOH values.

[0075] Then, draw from the prior distribution. Particles For each particle, instead of directly using the traditional state transition equation as the proposed distribution, an iterative unscented Kalman filter (IUKF) is run. The Sigma point is generated using the UT transform and combined with the real-time voltage obtained from the current measurement. The mean and covariance of the particles are updated through multiple iterations:

[0076]

[0077] This step pushes the particles to the high likelihood region, greatly mitigating particle degradation in the resampling step.

[0078] In actual processing, after extracting particles, the current SOC value and current SOH value from the previous round are usually used as prior data to calculate the particle weights. Then, the mean and covariance of the particles are updated based on the above processing method. The weights are iterated by RTS pre-estimation and Kalman prediction of the average value and variance. After each iteration, effective particles are selected according to their weights until the number of effective particles exceeds the threshold. Then, high-likelihood particles are output to complete the filtering process. The SOC value and SOH value corresponding to the particle are combined as the forward filtering estimate and output.

[0079] In one embodiment, such as Figure 5 As shown, step S4 includes: step S41: calculate the smoothing gain value for the forward filter estimate; step S42: perform reverse recursion on the forward filter estimate based on the smoothing gain value to correct fluctuations and obtain a smoothing correction result.

[0080] Specifically, to avoid data disturbance, this embodiment, after performing forward filtering, also predicts the future state of the SOC and SOH values ​​to achieve backward smoothing correction.

[0081] Specifically, after obtaining the forward filter estimate, the RTS smoothing process is initiated for reverse correction:

[0082] First, calculate the smoothing gain. :

[0083] ;

[0084] In the formula, For a moment The positively filtered posterior covariance reflects the initial confidence level of the forward estimation. It is determined by time Predicted time The prior prediction covariance matrix, The system state transition matrix describes the evolution of the battery state over time (derived by discretization of the second-order RC model).

[0085] The calculation of the smoothing gain essentially involves first deriving the system state transition matrix based on the discretized parameters of the previously established second-order RC model. Polarization parameters in a second-order RC circuit model Determines the matrix The off-diagonal elements in the equation, such as the coefficients of the polarization voltage components, are: Then, based on the system state transition matrix... The time was derived The prior prediction covariance matrix is ​​calculated, and the smoothing gain of the covariance matrix at two time points is calculated.

[0086] Then, reverse the recursive process to determine the state:

[0087] ;

[0088] In the formula, To utilize from arrive Information corrected for all observation times Status at any given moment. For the first Preliminary estimates of forward filtering at time step. To utilize the smoothed future state values ​​based on global information, For time-based For time The predicted state value.

[0089] Using data from future moments For the current moment The state is smoothed and corrected. This effectively filters out measurement noise generated by the test platform, reducing its impact on the slowly changing SOH calculation. Because... It is by This calculation means that if the physical parameters identified in the first stage are inaccurate, The matrix will become distorted, leading to incorrect predictions. It deviates from the actual physical state.

[0090] At this point, RTS passes smoothly. This approach astutely captures the residuals of the predicted and smoothed trajectories, and incorporates the smoothing gain. Reverse forced correction Transient jitter at any given moment.

[0091] This "global backtracking" mechanism can effectively filter out voltage pulses (glitch) caused by relay contact jitter, thereby ensuring that the SOC output curve maintains logical continuity and smoothness even under physical fault interference.

[0092] In one embodiment, such as Figure 6 As shown, step S5 includes: step S51: predicting the current SOC value based on the smoothing correction result; step S52: generating the current SOH value based on the current SOC value and the ohmic internal resistance in the battery model.

[0093] Specifically, after completing the smoothing correction, the corrected SOC value can be directly obtained as the current SOC value output. Subsequently, the ohmic internal resistance in the battery model obtained from the pre-updated parameters is combined with this value. Real-time SOH value updates:

[0094] ;

[0095] The updated model parameters are then fed back to the SOC estimator of the inner loop.

[0096] The above algorithm and existing algorithms were tested under a pulsed discharge environment, and the results were obtained. Figure 7a and Figure 8 The experimental results are shown.

[0097] Figure 7a This is a curve comparing the estimated and actual SOC values ​​generated during various algorithm tests under the influence of pulsed discharge. The magnified view A in the figure corresponds to... Figure 7b The enlarged view B corresponds to Figure 7c ; Figure 7b This is a magnified view of the SOC estimation and actual values ​​compared to the curves generated during the testing of various algorithms under the influence of pulse discharge, within the 0-500 second range. Figure 7c This is a magnified view of the SOC estimation and actual values ​​compared to the curves generated during the testing of various algorithms under the influence of pulse discharge, in the 3700-4200 second interval. Figure 8 This is a graph showing the absolute error of SOC generated during the testing of various algorithms under the influence of pulse discharge.

[0098] In pulsed discharge experiments conducted on the test platform, the traditional PF algorithm exhibited significant divergence in the later stages (due to particle depletion), with a mean absolute error of 1.24% for SOC. In contrast, the RTS-IUPF algorithm used in this embodiment consistently maintained a SOC error curve closely aligned with the true value, reducing the mean absolute error to 0.748%, and lowering the root mean square error (RMSE) by 77.8% compared to PF.1 In 500 cycles of aging testing, the SOH estimation error remained consistently below 4%, verifying the algorithm's adaptability throughout its entire lifecycle.

[0099] A battery testing system for implementing the above-described battery testing method; such as Figure 9 As shown, the battery testing system includes: a host computer A1, which connects to the battery management system of the battery under evaluation and measures the real-time current and voltage of the battery under evaluation; a battery cell simulator A2, which is controlled by the host computer and simulates the battery under evaluation based on the real-time current and voltage; and a fault injection circuit A3, which is connected to both the host computer A1 and the battery cell simulator A2. The fault injection circuit A3 is controlled by the host computer and injects faults into the battery cell simulator. The host computer A1 predicts the current SOC and current SOH values ​​of the battery under evaluation based on the response of the battery cell simulator A2 to the injected fault signal under simulated conditions.

[0100] Specifically, to verify the effectiveness of the above algorithm, this embodiment also constructs a specific test circuit to verify whether the algorithm can eliminate noise disturbances under fault conditions.

[0101] The battery testing system is connected to a battery to be evaluated. The host computer A1 obtains the actual physical parameters of the battery to be evaluated, including ohmic internal resistance and polarization internal resistance, through volt-ampere characteristic test or pulse discharge test. Then, these parameters are transmitted to the battery cell simulator A2. The battery cell simulator A2 is controlled by the host computer and simulates the battery to be evaluated according to the real-time current and real-time voltage, so that it has the corresponding physical parameters.

[0102] Based on this, the host computer A1 controls the battery cell simulator A2 to conduct charging and discharging experiments. During the charging and discharging process, the fault injection circuit A3 is controlled to generate corresponding fault signals.

[0103] Meanwhile, the aforementioned test algorithm is run in the host computer A1 to estimate the SOC and SOH values ​​of the battery cell simulator A2, in order to verify its effectiveness.

[0104] The fault injection circuit is implemented based on a high-speed switching array. It is connected in series between the battery cell simulator A2 and the sampling port of the host computer A1, and consists of relays. Unlike software parameter modification, this unit simulates "open circuit," "intermittent poor contact," and "short circuit" faults between channels through physical actions. The MCU normally keeps its I / O ports in input mode, switching to output mode only when a fault injection command is issued to prevent false triggering.

[0105] During testing, the control relay array was switched on and off at a frequency of 10Hz to simulate poor contact. The output of the RTS-IUPF algorithm was then observed. Due to the RTS smoothing structure, the SOC estimation result should remain smooth and should not jump due to instantaneous fluctuations in voltage sampling.

[0106] In one embodiment, such as Figure 10 As shown, it also includes an insulation detection circuit; the insulation detection circuit is set between the input and output terminals of the battery to be evaluated and is controlled by the host computer to test the insulation of the battery to be evaluated.

[0107] Specifically, to achieve better verification results, a voltage compensation circuit consisting of isolation capacitors (C198, C199, C200) and voltage divider resistors (R188, R189) was added to the traditional low-frequency pulse injection method. This circuit is connected between the positive and negative terminals (PACK+ / -) of the battery pack and can identify and automatically provide the compensation voltage required by the sampling circuit, ensuring that the output voltage ISO_AD can still quickly stabilize within the normal detection range under interference from large-capacity Y capacitors.

[0108] In this circuit, an external PWM excitation source is connected through the ISO_PWM port, and a buffer U123 is connected through the load resistor R111. Then, the signal is transmitted to the voltage divider network composed of voltage divider resistors R188 and R189 via resistors R112 and R113 to control the signal injected into the main and negative batteries. At the same time, the signal is filtered by isolation capacitors C198, C199 and C200 to prevent the signal from flowing back to the sampling port ISO_AD.

[0109] When an excitation is applied, the battery's main positive output response signal is filtered by isolation capacitors C199 and C200 and then transmitted to operational amplifier U133. Load resistors R116 and R115 are connected in series on this link for sampling, and the signal is input to the non-inverting input of operational amplifier U133 via load resistor R117. The inverting input of the operational amplifier is grounded or left floating, so that the signal is output to the sampling port ISO_AD via load resistor R119. This completes the testing process.

[0110] The circuit also includes other auxiliary devices, such as the power supply pin VCC of buffer U132, which is connected to the power supply circuit VCC, and the power supply circuit ripple is filtered out by grounding the filter capacitor C1.

[0111] For example, in the branch where the PWM excitation signal is input, a pair of Zener diodes are used to ground D11 and connect to the power supply circuit to control its input amplitude to be stable within the set range; similarly, in the sampling branch, a Zener diode is also used to ground D11 and connect to the power supply circuit to control the amplitude of the sampling signal output to the operational amplifier U133 to be stable within the set range.

[0112] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. A battery testing method based on adaptive state joint estimation, characterized in that, include: Step S1: Collect real-time voltage and real-time current of the battery to be evaluated, and establish a battery model corresponding to the resistive and capacitive characteristics of the battery to be evaluated based on the real-time voltage and real-time current. Step S2: Apply multiple innovation least squares method to the battery model to predict the predicted parameters of the battery model, and update the battery model based on the predicted parameters; Step S3: Extract particles based on the prior distribution, and run the unscented Kalman filter algorithm to iterate the particles according to the real-time voltage to obtain high-likelihood particles that match the real-time voltage as forward filter estimates. The particles represent different combinations of SOC and SOH values; Step S4: Perform reverse correction on the forward filter estimate to obtain a smooth correction result; Step S5: Based on the smoothing correction result, predict and output the current SOC value and the current SOH value; Step S2 includes: Step S21: Discretize the battery model to obtain a discretized model; Step S22: Calculate the gain matrix for the discretized model; In step S22, the method for calculating the gain matrix includes: Calculate the gain matrix based on the covariance matrix of the previous prediction: ； wherein is an information matrix consisting of data vectors from past time instants, is a covariance matrix of the last prediction, is the gain matrix; Step S23: Update the battery model according to the gain matrix to obtain parameter estimates as the prediction parameters; In step S23, the method for updating the battery model includes: ； In the formula, These are the parameter estimates obtained from this round of prediction. These are the prediction parameters obtained from the previous round of prediction. Here is the gain matrix. For a novel vector of predetermined length, the matrix of the prediction parameters corresponds to each parameter in the battery model; Step S24: Update the covariance matrix of the gain matrix to converge the gain matrix for the next update; Step S24 includes: ； In the formula, This is the covariance matrix for this round of predictions. This is the covariance matrix from the previous prediction. Here is the gain matrix. For the past An information matrix composed of data vectors at each time point.

2. The battery testing method of claim 1, wherein, The battery model in step S1 is based on a second-order RC model and includes: ； In the formula, The predicted terminal voltage of the battery to be evaluated. The open-circuit voltage corresponding to the current SOC value. The ohmic internal resistance of the battery to be evaluated is given. The polarization resistor of the first RC network. The polarization capacitor of the first RC network. The polarization voltage of the first RC network is denoted as . The polarization resistor of the second RC network, This is the polarization capacitance of the second RC network. is the polarization voltage of the second RC network.

3. The battery testing method of claim 1, wherein, Step S3 includes: Step S31: Extract the particles from the prior distribution, and generate a proposal distribution function for particle filtering based on iterative unscented Kalman filtering for the particles; Step S32: Iterate the proposed distribution function based on the real-time voltage to update the mean and covariance of the particles, and obtain the high likelihood particles.

4. The battery testing method of claim 2, wherein, Step S4 includes: Step S41: Calculate the smoothing gain value for the forward filter estimate; Step S42: Based on the smoothing gain value, the forward filter estimate is recursively calculated in reverse to correct for fluctuations and obtain the smoothing correction result.

5. The battery testing method of claim 2, wherein, Step S5 includes: Step S51: Predict the current SOC value based on the smoothing correction result; Step S52: Generate the current SOH value based on the current SOC value and the ohmic internal resistance in the battery model.

6. A battery testing system, comprising: Used to implement the battery testing method as described in any one of claims 1-5; The battery testing system includes: A host computer is connected to the battery management system of the battery to be evaluated and performs real-time current and real-time voltage measurements on the battery to be evaluated. A battery cell simulator, which is controlled by the host computer and simulates the battery to be evaluated based on the real-time current and the real-time voltage; A fault injection circuit is connected to the host computer and the battery cell simulator, respectively. The fault injection circuit is controlled by the host computer and performs fault injection on the battery cell simulator. The host computer predicts the current SOC and current SOH values ​​of the battery to be evaluated based on the response of the battery cell simulator to the injected fault signal under simulated conditions.

7. The battery testing system of claim 6, wherein, It also includes an insulation detection circuit; The insulation detection circuit is located between the input and output terminals of the battery to be evaluated and is controlled by the host computer to test the insulation of the battery to be evaluated.

8. The battery testing system of claim 6, wherein, The fault injection circuit is implemented based on a high-speed switch array.

Images