Active risk source identification method of energy storage battery based on Monte Carlo

Abstract

The invention relates to an active risk source identification method for an energy storage battery based on Monte Carlo, which comprises the following steps: re-sampling SOC (State of Charge) and SOH (State of Health) obtained by self-adaptive filtering estimation for multiple times by using a Monte Carlo method, averaging the results, regarding the average value as an optimal solution, and taking the average value as a final approximate estimation value at the time, so as to identify the risk source of the energy storage battery. And the influence of random noise is reduced. And comparing the obtained SOC and SOH parameters with nominal values, screening out abnormal batteries, performing fixed-point positioning early warning several weeks before short-circuit thermal runaway and normal aging failure in the batteries, obtaining high-precision internal parameters of the batteries through a sliding window recursive least square parameter identification method, and effectively suppressing noise through a Sage-Husa self-adaptive method. And meanwhile, the influence of inaccuracy of priori estimation on posteriori estimation is further reduced by using a Monte Carlo method. The battery state prediction precision is improved; the internal parameters of the battery can be identified on line, real-time parameter changes are obtained, robustness is good, and expandability is good.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage battery risk identification, and in particular to an active risk source identification method for energy storage batteries based on Monte Carlo. Background Art

[0002] With the continuous increase in the installed capacity of energy storage power stations, a large number of monomers are connected in series and parallel, which can easily lead to chain reactions and cause safety accidents such as fire and explosion. Therefore, its safety prevention and control issues are becoming more and more important. The active safety of power stations focuses on monitoring and evaluating the operating status of batteries during the operation of energy storage systems, identifying and warning early faults of the system, and then taking targeted maintenance measures to prevent serious accidents. Active safety technologies include but are not limited to battery management system (BMS) to evaluate the battery status and early fault warning. These technologies can intervene before the battery has problems and ensure the safe operation of the energy storage system. Among them, the active risk source identification of the battery is the most important link. By establishing an accurate battery model, real-time monitoring of the internal parameters of the battery, and estimating the battery state of charge (SOC) and health state (SOH) through filtering algorithms, the battery SOC and SOH parameters are compared with the nominal values, and the battery is monitored to see whether it can be charged and discharged normally, which reflects the safety status of the battery and screens out problematic batteries.

[0003] In practical applications, SOC and SOH are affected by the nonlinear characteristics of the battery and cannot be measured directly, making battery state estimation difficult and error-prone. They can only be estimated through measurable parameters such as voltage, current, and temperature. In addition, the battery system has inevitable random noise, and the estimation results are inaccurate under dynamic conditions. Therefore, improvement and innovation are imperative. Summary of the invention

[0004] In view of the above situation, in order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide an active risk source identification method for energy storage batteries based on Monte Carlo. Through the data extracted from the operating data, real-time online monitoring of the energy storage power station under operation status can be realized, and the state of the energy storage battery can be accurately predicted and judged, abnormal batteries can be screened out, and fixed-point positioning warnings can be made several weeks before the normal aging and failure of the battery.

[0005] The technical solution provided by the present invention is:

[0006] A Monte Carlo-based active risk source identification method for energy storage batteries comprises the following steps:

[0007] Step 1: Obtain the original operating data of the energy storage power station battery, including the battery terminal voltage and current of the energy storage power station battery;

[0008] Step 2: Online model parameter identification

[0009] (1) Establish a second-order RC equivalent circuit model, which includes resistances R0, R1, R2, and open circuit voltage U OC , capacitors C1 and C2; wherein, resistor R0 is the equivalent internal resistance of the battery model, resistor R1 is the polarization internal resistance caused by the battery polarization effect, resistor R2 is the polarization internal resistance caused by the battery concentration polarization effect, capacitors C1 and C2 are polarization capacitors, resistor R1 and capacitor C1 are connected in parallel to form the first RC network, the first RC network describes the impedance of transmission between electrodes, resistor R2 and capacitor C2 are connected in parallel to form the second RC network, the second RC network describes the impedance of lithium ion diffusion in the electrode material, the open circuit voltage U OC The positive electrode is connected in series with the resistor R0, the first RC network and the second RC network in sequence; U0 is the voltage divided by the resistor R0. When the current is I, the voltages divided by the resistors R1 and R2 are U1 and U2, U OC is the open circuit voltage, U L is the output terminal voltage; the differential expression of the second-order RC equivalent circuit model is:

[0010]

[0011] Where: U1 represents the voltage across R1, U2 represents the voltage across R2, are the derivatives of U1 and U2, respectively. L Indicates the circuit terminal voltage, U oc represents the open circuit voltage, R0 represents the internal resistance of the battery, R1 represents the first electrochemical polarization resistance, R2 represents the second concentration polarization resistance, C1 represents the first electrochemical polarization capacitance, C2 represents the second concentration polarization capacitance, and I represents the current;

[0012] (2) Discretize the established second-order RC equivalent circuit model expression to obtain the following expression:

[0013]

[0014] Where: k represents the discrete moment, Δt represents the sampling time, τ represents the time constant, and τ1=R1C1, τ2=R1C2, C represents the rated capacity of the target battery;

[0015] The expression of the discretized second-order RC equivalent circuit model is sorted out to obtain the following expression:

[0016]

[0017] Performing Laplace transform on the expression of the sorted second-order RC equivalent circuit model, we get the following expression:

[0018]

[0019] Will U OC (s)-U L (s) as output and I(s) as input, the transfer function expression is obtained:

[0020]

[0021] make Among them, the transformed transfer function is discretized to obtain the following expression:

[0022] Among them, ε1, ε2, ε3, ε4, ε5 are constant coefficients;

[0023] Substituting the transfer function G(z) into the expression of the second-order RC equivalent circuit model after Laplace transformation, the following expression is obtained:

[0024] y θ (k)=ε1y θ (k-1)+ε2y θ (k-2)+ε3I(k)+ε4I(k-1)+ε5I(k-2)

[0025] in:

[0026]

[0027] Let θ(k)={ε1, ε2, ε3, ε4, ε5}, Then the above expression can be written as: θ (k)=φ(k) T θ(k), where y θ (k) represents the output vector at time k, φ(k) represents the input vector at time k, and θ(k) represents the model parameters to be identified at time k;

[0028] (3) Input voltage and current, and recursively calculate the internal parameters of the equivalent circuit model through the SWRLS algorithm

[0029] The sliding window length is set to W. By recursively calculating the W+1 sets of data between time kW and time k and the data at time k-1 as the initial value, the system state at time k can be obtained. By replacing kW with k-W+1 in the equation, the SWRLS algorithm iteration is shown as follows:

[0030]

[0031] in, is the estimated value of the identified parameter at time k after iteration when the window slides; is the estimated value of the identified parameter at time k before iteration when the window slides; G(k-W+1,k) is the algorithm gain matrix when the window slides; y(kW) is the state variable at time k when setting the window; φ(kW) is the parameter to be identified at time k; P(k-W+1,k) is the covariance matrix at time k after iteration when the window slides; P(kW,k) is the covariance matrix at time k before iteration when setting the window;

[0032] Step 3: Battery Status Estimation

[0033] (1) Based on the expression of the discretized second-order RC equivalent circuit model, the corresponding state space equation and observation equation are determined:

[0034]

[0035] Where x(k-1), x(k) represents the state vector at time k-1 and time k, u(k-1) represents the input vector at time k-1, y(k) represents the observation vector at time k, A represents the system state transfer matrix, B represents the system state control matrix, C represents the observation transfer matrix, D represents the observation control matrix, w(k) represents the process noise vector at time k, and v(k) represents the measurement noise vector at time k.

[0036]

[0037] D=-R0

[0038] (2) Initialize state variables Error covariance matrix P k , noise matrix Q k , R k ,q k 、r k ;

[0039] (3) Input the battery terminal voltage, current, and internal parameters of the equivalent circuit model;

[0040] (4) Calculate the noise matrix according to the Sage-Husa adaptive filter:

[0041]

[0042] (5) Pre-estimation of state variables and state error covariance:

[0043] State variable estimation State Error Covariance Pre-estimation

[0044] (6) Kalman gain update:

[0045]

[0046] (7) Update the observed state measurement value, state variable, and state error covariance:

[0047] Update observation status measurement value State variables State Error Covariance

[0048] (8) Monte Carlo process:

[0049] Resample the state variables to obtain the optimal estimate Output the SOC and SOH of the energy storage battery.

[0050] The innovation of the invention method lies in applying the Monte Carlo mathematical statistics method to the field of battery state estimation. It uses the Monte Carlo method to resample the SOC and SOH obtained by adaptive filtering estimation multiple times, and averages the results. The average value is considered to be the optimal solution, and is used as the final approximate estimate at that time, reducing the impact of random noise. The obtained SOC and SOH parameters are compared with the nominal values ​​to screen out abnormal batteries, and a fixed-point positioning warning is made several weeks before short-circuit thermal runaway and normal aging failure in the battery. Compared with the prior art, the advantages of this application are as follows:

[0051] (1) High accuracy: The sliding window recursive least squares parameter identification method is used to obtain high-precision battery internal parameters, and the Sage-Husa adaptive method is used to effectively suppress noise. At the same time, the Monte Carlo method is used to further reduce the impact of the inaccuracy of the prior estimation on the posterior estimation, thereby improving the accuracy of battery state prediction.

[0052] (2) Strong real-time performance: Compared with other parameter identification methods, the sliding window recursive least squares parameter identification method is more sensitive to new data. It can identify the internal parameters of the battery online and obtain real-time parameter changes, providing a basis for battery state estimation.

[0053] (3) Good robustness: The adaptive correction method effectively suppresses noise interference and improves the robustness and stability of battery state prediction.

[0054] (4) Good scalability: It has good scalability and versatility, can be applied to battery status monitoring applications in different scenarios, and provides a method for active risk source identification of energy storage batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a second-order RC equivalent circuit model diagram of the present invention;

[0056] Figure 2 The figure is a flow chart of the algorithm of the present invention.

[0057] Figure 3 It is the overall operation flow chart of the present invention.

[0058] Figure 4 This is a screenshot of the actual measurement of the "battery management system" after the application of the present invention. DETAILED DESCRIPTION

[0059] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings.

[0060] The present invention combines the Monte Carlo mathematical statistics method (Monte Carlo) with the Sage-Husa adaptive filter and the dual extended Kalman filter (DEKF) algorithm, and proposes a (Monte Carlo-Sage Husa-Dual extended Kalman filter, MC-SH-DEKF) algorithm to realize the active risk source identification of energy storage batteries. The established battery equivalent circuit model is parameter identified by the sliding window recursive least squares (SWRLS), and then the identified battery parameters are passed to the Sage-Husa adaptive filter, and the SOC and SOH are obtained by combining the DEKF iteration. The SOC and SOH estimated by the adaptive filter are resampled multiple times using the Monte Carlo method to obtain the optimal solution of SOC and SOH at that time, and compared with their nominal values, abnormal batteries are screened out, and a fixed-point positioning warning is made several weeks before the battery aging and failure.

[0061] Specifically, the objectives of this approach include:

[0062] 1. Battery parameter identification: By extracting and analyzing the operating data of the energy storage power station battery, an equivalent model of the battery is established, and the internal parameters of the battery, such as internal resistance and capacitance changes, are identified through the sliding window recursive least squares method.

[0063] 2. System performance monitoring: Through real-time online evaluation of battery safety status based on adaptive filtering and Monte Carlo methods, a comprehensive and accurate dynamic evaluation is provided for the full life cycle operation performance evaluation of energy storage power stations.

[0064] 3. Risk source warning: Compare the SOC and SOH parameters of the obtained battery with the nominal values ​​to screen out abnormal batteries.

[0065] Provide pinpoint positioning warning weeks before normal battery aging failure.

[0066] In summary, the purpose of the present invention is to extract the battery operation data of the energy storage power station, identify the internal parameters of the battery, evaluate the battery status, issue early warnings for abnormal batteries, ensure the safe and stable operation of the energy storage system, and improve the efficiency of the energy storage system.

[0067] Therefore, the present invention provides an active risk source identification method for energy storage batteries based on Monte Carlo, comprising the following steps:

[0068] Step 1: Obtain the original operating data of the energy storage power station battery, including the battery terminal voltage and current of the energy storage power station battery;

[0069] Step 2: Online model parameter identification

[0070] (1) Establish a second-order RC equivalent circuit model,

[0071] It includes resistors R0, R1, R2, open circuit voltage U OC , capacitors C1 and C2; wherein, resistor R0 is the equivalent internal resistance of the battery model, resistor R1 is the polarization internal resistance caused by the battery polarization effect, resistor R2 is the polarization internal resistance caused by the battery concentration polarization effect, capacitors C1 and C2 are polarization capacitors, resistor R1 and capacitor C1 are connected in parallel to form the first RC network, the first RC network describes the impedance of transmission between electrodes, resistor R2 and capacitor C2 are connected in parallel to form the second RC network, the second RC network describes the impedance of lithium ion diffusion in the electrode material, the open circuit voltage U OC The positive electrode is connected in series with the resistor R0, the first RC network and the second RC network in sequence; U0 is the voltage divided by the resistor R0. When the current is I, the voltages divided by the resistors R1 and R2 are U1 and U2, U OC is the open circuit voltage, U L is the output terminal voltage;

[0072] The present invention establishes a second-order RC model as an equivalent circuit model of a battery. It can simulate concentration polarization and electrochemical polarization respectively, so as to accurately simulate the dynamic characteristics of the battery. Compared with other equivalent models, it can better characterize the time-varying characteristics of the internal parameters of the lithium battery, and control the complexity of the model within a reasonable range while satisfying the polarization capacitance and polarization resistance characteristics.

[0073] The second-order RC equivalent circuit model adds a resistance-capacitance (RC) loop on the basis of the Rint model to characterize the polarization effect of power lithium-ion batteries, which solves the error caused by the Rint model not considering the polarization effect of power lithium-ion batteries. Therefore, it has higher accuracy and can effectively characterize the working characteristics of the battery. This model can better characterize the dynamic response of the battery. The ohmic resistance can represent the instantaneous change of the voltage response of the battery during charging and discharging, and the RC loop can reflect the gradual change of the battery voltage during and after charging and discharging. The second-order RC equivalent circuit model is shown in Appendix 1 Figure 1 shown.

[0074] Figure 1 In the figure, R0 is the equivalent internal resistance of the battery model, R1 is the polarization internal resistance caused by the battery polarization effect, R2 is the polarization internal resistance caused by the battery concentration polarization effect, C1 and C2 are polarization capacitors. The first RC network describes the impedance of transmission between electrodes, the second RC network describes the impedance of lithium ion diffusion in the electrode material, and U0 is the voltage divided by R0. When the current is I, the voltage divided by resistors R1 and R2 is U1 and U2. U OC is the open circuit voltage, U L is the output terminal voltage.

[0075] According to Kirchhoff's voltage and current laws, the voltage and current of the equivalent circuit are shown in equation (1).

[0076]

[0077] The advantages of the second-order RC equivalent circuit model include: ① For the nonlinear characteristics of energy storage lithium batteries, it can well simulate the dynamic response during operation; ② The model parameters are easy to identify and have practical physical meanings; ③ When using this model, the circuit equation can be easily converted into a state space equation. At present, the second-order RC equivalent circuit model is also widely used in practice. It has been widely used in the research of accurate SOC estimation and has provided a reference for the development of more suitable models.

[0078] The differential expression of the second-order RC equivalent circuit model is:

[0079]

[0080] Where: U1 represents the voltage across R1, U2 represents the voltage across R2, are the derivatives of U1 and U2, respectively. L Indicates the circuit terminal voltage, U OCrepresents the open circuit voltage, R0 represents the internal resistance of the battery, R1 represents the first electrochemical polarization resistance, R2 represents the second concentration polarization resistance, C1 represents the first electrochemical polarization capacitance, C2 represents the second concentration polarization capacitance, and I represents the current;

[0081] (2) Discretize the established second-order RC equivalent circuit model expression to obtain the following expression:

[0082]

[0083] Where: k represents the discrete moment, Δt represents the sampling time, τ represents the time constant, and τ1=R1C1, τ2=R1C2, C represents the rated capacity of the target battery;

[0084] The expression of the discretized second-order RC equivalent circuit model is sorted out to obtain the following expression:

[0085]

[0086] Performing Laplace transform on the expression of the sorted second-order RC equivalent circuit model, we get the following expression:

[0087]

[0088] Where: s is the complex variable in the Laplace transform, which is used to convert the differential equation operation in the time domain into an algebraic operation;

[0089] Will U OC (s)-U L (s) as output and I(s) as input, the transfer function expression is obtained:

[0090]

[0091] make In Laplace transform, negative powers of z (z -1 ) is used to represent the transfer function, T is the frequency domain expression of the time domain Δt in the Laplace transform, and the transformed transfer function is discretized to obtain the following expression:

[0092] Among them, ε1, ε2, ε3, ε4, ε5 are constant coefficients;

[0093] Substituting the transfer function G(z) into the expression of the second-order RC equivalent circuit model after Laplace transformation, the following expression is obtained:

[0094] y θ (k)=ε1y θ (k-1)+ε2y θ(k-2)+ε3I(k)+ε4I(k-1)+ε5I(k-2)

[0095] in:

[0096]

[0097] Let θ(k)={ε1, ε2, ε3, ε4, ε5}, Then the above expression can be written as: θ (k)=φ(k) T θ(k), where y θ (k) represents the output vector at time k, φ(k) represents the input vector at time k, and θ(k) represents the model parameters to be identified at time k;

[0098] (3) Input voltage and current, and recursively calculate the internal parameters of the equivalent circuit model through the SWRLS algorithm

[0099] The sliding window length is set to W. By recursively calculating the W+1 groups of data between time kW and time k and the data at time k-1 as the initial value, the system state at time k can be obtained. Since the sliding window length W should remain unchanged, the old data at kW should be deleted when reading the new data at time k. Therefore, by replacing kW with k-W+1 in the equation, the SWRLS algorithm iteration is as follows:

[0100]

[0101] in, is the estimated value of the identified parameter at time k after iteration when the window slides; is the estimated value of the identified parameter at time k before iteration when the window slides; G(k-W+1,k) is the algorithm gain matrix when the window slides; y(kW) is the state variable at time k when setting the window; φ(kW) is the parameter to be identified at time k; P(k-W+1,k) is the covariance matrix at time k after iteration when the window slides; P(kW,k) is the covariance matrix at time k before iteration when setting the window;

[0102] The present invention proposes a parameter identification method based on sliding window recursive least squares (SWRLS), which can further solve the data saturation problem that may exist in the forgetting factor recursive least squares algorithm. The traditional forgetting factor recursive least squares (FFRLS) multiplies the data by a weight according to the degree of newness of the data to reduce the influence of the old data. The SWRLS method directly deletes the old data when a new set of data is input, and only uses the latest data within the sliding window length for parameter estimation. This enables the algorithm to obtain more information from the new data and has better recognition performance.

[0103] Step 3: Battery Status Estimation

[0104] (1) Based on the expression of the discretized second-order RC equivalent circuit model, the corresponding state space equation and observation equation are determined:

[0105]

[0106] Where x(k-1), x(k) represents the state vector at time k-1 and time k, u(k-1) represents the input vector at time k-1, y(k) represents the observation vector at time k, A represents the system state transfer matrix, B represents the system state control matrix, C represents the observation transfer matrix, D represents the observation control matrix, w(k) represents the process noise vector at time k, and v(k) represents the measurement noise vector at time k.

[0107]

[0108] D=-R0

[0109] (2) Initialize state variables Error covariance matrix P k , noise matrix Q k , R k ,q k 、r k ;

[0110] (3) Input the battery terminal voltage, current, and internal parameters of the equivalent circuit model;

[0111] (4) Calculate the noise matrix according to the Sage-Husa adaptive filter:

[0112] In order to reduce the error caused by the extended Kalman filter algorithm ignoring the influence of noise in the actual process, the Sage-Husa adaptive filter is used to estimate and correct the process and measurement noise of the system by comparing the final estimation result with the prediction result. At the same time, the process noise and measurement noise of the system are estimated and corrected, and the average estimate and covariance are adjusted to reduce the influence of noise on the charge state estimation, thereby obtaining a higher accuracy of the charge state estimation result. The calculation process of the noise variable of the Sage-Husa adaptive filter is shown in the following formula:

[0113]

[0114] in, is the system state variable at time k, y k+1 is the measured value of the state observation at time k+1, G is the noise-driven matrix, is the error covariance matrix of the initial forecast, d k is the weighting coefficient at time k; b is the forgetting factor, which is 0.98 here, Q k is the covariance matrix of the system state noise at time k, R k is the covariance matrix of the system observation noise at time k, q k is the system state noise at time k, r k is the system observation noise at time k, Q k+1 is the covariance matrix of the system state noise at time k+1, R k+1 is the covariance matrix of the system observation noise at time k+1, q k+1 is the system state noise at time k+1, r k+1 is the system observation noise at time k+1, is the initial value of the state observation at time k+1, is the estimated value of the input vector at time k, u k+1 is the input vector at time k+1, is the system state variable at time k+1, K k+1 is the Kalman gain at time k+1, the superscript T represents matrix transpose, A represents the system state transfer matrix, B represents the system state control matrix, C represents the observation transfer matrix, and D represents the observation control matrix.

[0115] (5) Pre-estimation of state variables and state error covariance:

[0116] State variable estimation State Error Covariance Pre-estimation

[0117] (9) Kalman gain update:

[0118]

[0119] (10) Update the observed state measurement value, state variable, and state error covariance:

[0120] Update observation status measurement value State variables State Error Covariance

[0121] (11) Monte Carlo process:

[0122] The Monte Carlo method is a numerical simulation method that studies probabilistic phenomena. It is a calculation method that estimates unknown characteristic quantities by obtaining statistical values ​​based on sampling survey methods. It is suitable for the calculation and simulation of discrete systems. The basic method of Monte Carlo is to estimate the probability through a large number of repeated tests and statistical frequencies to obtain the solution to the problem. Due to the existence of noise in the system, the noise is reduced through an adaptive process. In the calculation process, the SOC value is related to the noise. The actual SOC is resampled many times through the Monte Carlo method, and the results are averaged. SOC is considered to be the optimal solution, that is, the real value of SOC. In the SOC estimation process, in order to reduce the impact of random noise on the estimation results, the SOC is sampled 1000 times and estimated repeatedly each time. Finally, the average value of 1000 samples each time is calculated as the final approximate estimate of the SOC at that time. The calculation formula for the approximate estimate of SOC each time is shown as follows:

[0123]

[0124] In the formula It is the estimated value of the state variable at time k. The SOH calculation is similar, and the SOC and SOH of the energy storage battery can be output.

[0125] Since the algorithm takes into account the influence of real noise, the Sage-Husa adaptive filter estimates its noise, and the Monte Carlo method can further reduce the influence of random noise on the estimation results. At the same time, it also reduces the influence of the inaccuracy of the prior estimate on the posterior estimate.

[0126] The calculation of SOC and SOH is shown in the following formula:

[0127]

[0128] SOC k0 is the SOC value at time k0; η is the Coulomb coefficient; Q0 is the factory capacity of the battery; Q n,k is the battery capacity at time k.

[0129] SOC represents the ratio of the available capacity of the battery to the rated capacity, and the available capacity is related to the operating state that represents the degree of battery aging. As the battery charge and discharge cycles increase, the available capacity of the battery gradually decreases. Therefore, DEKF is used here to jointly estimate the battery SOC and SOH. One EKF is used for SOC estimation, and the other EKF is used for capacity estimation.

[0130] The method of the present invention applies the Monte Carlo mathematical statistics method to the field of battery state estimation. Since SOC and SOH are affected by the nonlinear characteristics of the battery, and the battery system has inevitable random noise, the estimation results are inaccurate under dynamic conditions. Therefore, the Sage-Husa adaptive filter and the DEKF algorithm are combined to reduce the influence of random noise and estimate the battery SOC and SOH. At the same time, the SOC obtained by adaptive filtering estimation is resampled multiple times using the Monte Carlo method, and the results are averaged to obtain the optimal SOC value, thereby obtaining the SOH at that time. The obtained SOC and SOH parameters are compared with the nominal values ​​to screen out abnormal batteries, and a fixed-point positioning warning is made several weeks before short-circuit thermal runaway and normal aging failure in the battery. In summary, the present invention proposes a (Monte Carlo-Sage Husa-Dual extended Kalman filter, MC-SH-DEKF) algorithm to realize active risk source identification of energy storage batteries.

[0131] The present invention has achieved good technical effects through practical application, and the application examples are as follows:

[0132] A certain energy storage power station has currently deployed a battery management system (BMS) developed using the method of the present invention to monitor the internal conditions of the battery in real time. Figure 4 shown.

[0133] As shown in the figure, the system detects abnormal attenuation of the available capacity of the battery cluster, issues an early warning, and analyzes the fault and proposes a solution. Cause 1: The capacity attenuation rate of individual battery cells is too large; Cause 2: The internal resistance of individual battery cells is obviously abnormal; Cause 3: The temperature difference between battery cells is relatively large. Solution: 1. Check whether the health of each module and cluster in the same battery compartment is abnormal, and if it occurs continuously, perform charging and discharging maintenance; 2. Perform capacity calibration when the battery is fully charged; 3. Check the problem battery cells.

[0134] By accurately monitoring and managing the battery charging and discharging process, the safety, stability and efficiency of the battery pack are ensured. The application of battery management system in energy storage system includes battery status monitoring, charge status and health status estimation, safety protection mechanism, etc. The monitored battery parameters mainly include single cell voltage, single cell temperature, total current, etc. High requirements are placed on the accuracy and timeliness of the parameters collected by the battery management system, because it determines the stability of the entire system. The detection of single cell voltage is also to avoid the occurrence of single cell over-voltage and under-voltage. In the battery management system of the actual application of energy storage power station, the voltage of each single cell needs to be collected. One of the functions of this platform is to estimate the internal SOC and SOH of the battery, compare the obtained SOC and SOH parameters with the nominal values, screen out abnormal batteries, and make fixed-point positioning warnings several weeks before the normal aging and failure of the battery. Different battery management systems have different fault alarm levels, which are divided into 3 categories: level 1 warning, that is, serious warning, the relay needs to be cut off immediately; level 2 warning, that is, moderate warning, standby, charging and discharging are prohibited; level 3 warning, that is, minor fault, request power reduction. The battery management system will make corresponding handling measures according to different faults and fault levels.

Claims

1. A Monte Carlo-based active risk source identification method for energy storage batteries, characterized in that: The following steps are involved: Step 1: Obtain the original operating data of the energy storage power station battery, including the battery terminal voltage and current of the energy storage power station battery; Step 2: Online model parameter identification (1) Establish a second-order RC equivalent circuit model, which includes resistances R0, R1, R2, and open circuit voltage U OC , capacitors C1, C2; Among them, the resistor R0 is the equivalent internal resistance of the battery model, the resistor R1 is the polarization internal resistance caused by the battery polarization effect, the resistor R2 is the polarization internal resistance caused by the battery concentration polarization effect, the capacitors C1 and C2 are polarization capacitors, the resistor R1 and the capacitor C1 are connected in parallel to form the first RC network, the first RC network describes the impedance of transmission between the electrodes, the resistor R2 and the capacitor C2 are connected in parallel to form the second RC network, the second RC network describes the impedance of lithium ion diffusion in the electrode material, the open circuit voltage U OC The positive electrode is connected in series with the resistor R0, the first RC network and the second RC network in sequence; U0 is the voltage divided by the resistor R0. When the current is I, the voltages divided by the resistors R1 and R2 are U1 and U2, U OC is the open circuit voltage, U L is the output terminal voltage; the differential expression of the second-order RC equivalent circuit model is: Where: U1 represents the voltage across R1, U2 represents the voltage across R2, are the derivatives of U1 and U2, respectively. L Indicates the circuit terminal voltage, U oc represents the open circuit voltage, R0 represents the internal resistance of the battery, R1 represents the first electrochemical polarization resistance, R2 represents the second concentration polarization resistance, C1 represents the first electrochemical polarization capacitance, C2 represents the second concentration polarization capacitance, and I represents the current; (2) Discretize the established second-order RC equivalent circuit model expression to obtain the following expression: Where: k represents the discrete moment, Δt represents the sampling time, τ represents the time constant, and τ1=R1C1, τ2=R1C2, C represents the rated capacity of the target battery; The expression of the discretized second-order RC equivalent circuit model is sorted out to obtain the following expression: Performing Laplace transform on the expression of the sorted second-order RC equivalent circuit model, we get the following expression: Will U OC (s)-U L (s) as output, I(s) as input, and the transfer function expression is obtained: make The transformed transfer function is discretized to obtain the following expression: Among them, ε1, ε2, ε3, ε4, ε5 are constant coefficients; Substituting the transfer function G(z) into the expression of the second-order RC equivalent circuit model after Laplace transformation, the following expression is obtained: y θ (k)=ε1y θ (k-1)+ε2y θ (k-2)+ε3I(k)+ε4I(k-1)+ε5I(k-2) in: Let θ(k)={ε1, ε2, ε3, ε4, ε5}, Then the above expression can be written as: θ (k)=φ(k) T θ(k), where y θ (k) represents the output vector at time k, φ(k) represents the input vector at time k, and θ(k) represents the model parameters to be identified at time k; (3) Input voltage and current, and recursively calculate the internal parameters of the equivalent circuit model through the SWRLS algorithm The sliding window length is set to W. By recursively calculating the W+1 groups of data between time kW and time k and the data at time k-1 as the initial value, the system state at time k can be obtained. The SWRLS algorithm iteration is shown as follows: in, is the estimated value of the identified parameter at time k after iteration when the window slides; is the estimated value of the identified parameter at time k before iteration when the window slides; G(k-W+1,k) is the algorithm gain matrix when the window slides; y(kW) is the state variable at time k when setting the window; φ(kW) is the parameter to be identified at time k; P(k-W+1,k) is the covariance matrix at time k after iteration when the window slides; P(kW,k) is the covariance matrix at time k before iteration when setting the window; Step 3: Battery Status Estimation (1) Based on the expression of the discretized second-order RC equivalent circuit model, the corresponding state space equation and observation equation are determined: Where x(k-1), x(k) represents the state vector at time k-1 and time k, u(k-1) represents the input vector at time k-1, y(k) represents the observation vector at time k, A represents the system state transfer matrix, B represents the system state control matrix, C represents the observation transfer matrix, D represents the observation control matrix, w(k) represents the process noise vector at time k, and v(k) represents the measurement noise vector at time k. D=-R0 (2) Initialize state variables Error covariance matrix P k , noise matrix Q k , R k ,q k 、r k ; (3) Input the battery terminal voltage, current, and internal parameters of the equivalent circuit model; (4) Calculate the noise matrix according to the Sage-Husa adaptive filter: (5) Pre-estimation of state variables and state error covariance: State variable estimation State Error Covariance Pre-estimation (6) Kalman gain update: (7) Update the observed state measurement value, state variable, and state error covariance: Update observation status measurement value State variables State Error Covariance (8) Monte Carlo process: Resample the state variables to obtain the optimal estimate Output the SOC and SOH of the energy storage battery.

Images