Online estimation method of lithium battery SOP based on RNN neural network and multi-parameter constraints

Abstract

The present invention discloses a method for online real-time estimation of lithium-ion battery SOP based on RNN neural network and multi-parameter constraints, and online identification of each parameter in the circuit model is performed based on UDDS working conditions, and an extended Kalman filter method is used to complete the online estimation of SOC under cyclic working conditions. Based on the constraints of the continuous peak current of the lithium battery SOC and the circuit model, the multi-parameter constraint real-time estimation of the lithium battery SOP is completed using the RNN neural network. The present invention estimates through the RNN neural network model, adds multi-parameter constraints, makes the lithium battery SOP deviation smaller, and considers multiple constraints to estimate the battery SOP at the same time, thereby improving the estimation accuracy of the SOP.

Description

Technical Field

[0001] The invention belongs to the field of electric vehicle battery management, and in particular relates to a lithium battery SOP online estimation method. Background Art

[0002] With the continuous development of the automobile industry, the contradictions of energy consumption and environmental pollution caused by traditional fuel vehicles have gradually emerged and become factors that must be considered in social progress. Therefore, increasing the development of electric vehicles, reducing the cost of using electric vehicles, and implementing a low-carbon and sustainable development strategy can alleviate people's current concerns about energy and environmental issues.

[0003] The power state (SOP) of a lithium battery generally refers to the continuous peak power state, which indicates the continuous peak discharge or charging power at different charge states over a period of time under the premise of ensuring the normal operation of the battery. Accurately estimating the SOP is important for ensuring that the braking energy can be recovered to the maximum during braking of an electric vehicle, providing as much power as possible during acceleration or starting, and shortening the time required to fully charge the battery in fast charging mode.

[0004] The main methods for estimating SOP include interpolation method, model method and data-driven method. The interpolation method based on HPPC test often requires a large number of experiments to test the charging and discharging power value after the battery voltage reaches the limited voltage. Although the principle is simple, this method ignores the polarization phenomenon of the battery and has poor dynamic characteristics. The model method is divided into a battery model-based method and a SOC-based method. The battery model-based method can well show the nonlinear characteristics of lithium battery SOP, but different models will affect the estimation accuracy of SOP. Using the SOC-based method alone will cause the estimated peak current to be too large, and there will be safety hazards. The RNN neural network data-driven method has more accurate input and output and has a high reference value. Summary of the invention

[0005] In order to solve the technical problems mentioned in the above background technology, the present invention proposes an online estimation method for SOP of lithium batteries based on RNN neural network and multi-parameter constraints.

[0006] In order to achieve the above technical objectives, the technical solution of the present invention is:

[0007] A lithium battery SOP online estimation method based on RNN neural network and multi-parameter constraints includes the following steps:

[0008] (1) Establishing a lithium battery equivalent circuit model and dynamic system equation, performing lithium battery OCV testing, recording lithium battery data under UDDS conditions, and performing online identification of various parameters of the lithium battery using an extended Kalman filter algorithm; the parameters include internal resistance R0, polarization resistance, and polarization capacitance;

[0009] (2) The EKF discrete nonlinear system equation is established based on the equivalent circuit model of the lithium battery. The state equation and observation equation of the lithium battery are established through the extended Kalman filter to complete the online estimation of SOC under the cycle condition.

[0010] (3) Estimate continuous peak current and continuous peak power;

[0011] (4) Taking the terminal voltage, current and SOC under the battery working condition as input and the peak power SOP as output, the RNN neural network model is imported, the RNN neural network is established, and the peak power of the lithium battery is estimated.

[0012] Further, in step (1), the lithium battery equivalent circuit model is a second-order RC circuit model, and the second-order RC circuit model includes an electrochemical polarization internal resistance R1, an electrochemical polarization capacitor C1, a concentration polarization resistor R2, and a concentration polarization capacitor C2;

[0013] The dynamic system equation is as follows:

[0014] U0=U oc (t)-R0·I(t)-U1(t)-U2(t)

[0015] Among them, U0 represents the terminal voltage of the battery in working state, U oc (t) is the expression of open circuit voltage, R0 is the expression of ohmic internal resistance, I is the discharge current, U1 and U2 are respectively represented by two sets of RC loop terminal voltages, and t is time.

[0016] Furthermore, in step (1), an OCV test is performed on a 18650 lithium-ion battery at room temperature, and the test data is fitted to obtain a corresponding relationship curve between open circuit voltage and SOC.

[0017] Furthermore, the specific process of step (1) is as follows:

[0018] (11) Complete the UDDS operating condition test of the lithium battery at room temperature and record the voltage, current and SOC true value;

[0019] (12) Storing intermediate process estimates;

[0020] (13) Initialize the parameters and convert the estimated values ​​of the parameters into R0, R1, R2, C1, C2; update the values ​​at the next moment;

[0021] (14) Calculate the Kalman filter gain matrix and collect the voltage and current data at the next moment;

[0022] (15) Repeat (12) to (14) to identify the lithium battery model parameters online.

[0023] Furthermore, the specific process of step (2) is as follows:

[0024] (21) Establish the EKF discrete nonlinear system equation:

[0025] χ k+1 =f(x k ,u k )+w k

[0026] y k =g(x k ,u k )+v k

[0027] Among them, χ k+1 is the input, y k is the output, w k and v k is the noise of the system and follows a normal distribution, which is similar to f(x k ,u k ), g(x k ,u k ) are independent of each other, f(x k ,u k ) and g(x k ,u k ) is a nonlinear function;

[0028] (22) The state equation and observation equation of lithium battery are established by the extended Kalman filter algorithm:

[0029]

[0030] U k =U oc (s k )-i k R0-U 1,k -U 2,k

[0031] Among them, τ d and τ e are two different time constants, U oc is the open circuit voltage, s k is the estimated SOC value at time k, i k is the current at time k, T is the sampling time, C N is the rated capacity of the battery, η tis the charge and discharge efficiency, U k is the total estimated value of the terminal voltage at time k, U 1,k is the estimated value of the terminal voltage on R1 at time k, U 2,k is the estimated value of the terminal voltage on R2 at time k;

[0032] (23) Based on the terminal voltage and current data collected during the charging and discharging process of the lithium battery, the discretized state equation and observation equation are subjected to EKF recursive calculation to complete the online estimation of SOC under cyclic conditions.

[0033] Furthermore, the specific process of step (3) is as follows:

[0034] (31) Estimate the peak current based on the battery SOC:

[0035] Assume that the battery has a current i per unit time Δt. k Discharging, the battery SOC at time t+Δt is as follows:

[0036]

[0037] When the battery is working, the battery SOC value at any time satisfies the following constraints: SOC min <SOC(t)<SOC max , then the maximum charge and discharge current under SOC constraint can be deduced from the above formula:

[0038]

[0039] Where η i is the battery coulombic efficiency, Q v is the actual capacity of the battery, SOC(t) is the SOC value at time t, SOC min , SOC max are the minimum and maximum values ​​of SOC, η chg and η dis For battery charging and discharging efficiency, is the minimum charging current based on the battery SOC, is the maximum discharge current based on the battery SOC;

[0040] (32) Based on the second-order RC model of the battery, estimate the peak current:

[0041] According to the terminal voltage constraint: U t,min t t,max , the maximum charge and discharge current estimate is:

[0042]

[0043] In the formula, and​​ are the estimates of the peak discharge and charge currents based on the second-order RC model of the battery at time k, L is the continuous sampling time; g(SOC) represents the open circuit voltage of the battery, SOC k+1 Indicates the battery SOC and current i at time k+1 k+1 Functional relationship of It means the first-order partial derivative of g(SOC) with respect to SOC;

[0044] (33) Estimate the peak power of lithium battery based on the multi-parameter constraints of battery SOC and battery second-order RC model:

[0045] First, estimate the peak current under multi-parameter constraints:

[0046]

[0047] Among them, i max 、i min are the preset maximum discharge current and minimum charge current, and are respectively the maximum value of the maximum discharge current and the minimum value of the minimum charge current based on the battery SOC within the continuous sampling time L;

[0048] Then estimate the peak power:

[0049]

[0050] Among them, C N is the rated capacity of the battery, j represents the initial moment in the continuous sampling time, τ is the time constant, is the maximum discharge peak power and minimum charge peak power under multi-parameter constraints, g(SOC,C N ) represents the open circuit voltage of the battery based on the rated capacity of the battery, represents g(SOC,C N )Calculate the first-order partial derivative of SOC.

[0051] Furthermore, the specific process of establishing the RNN neural network in step (4) is as follows:

[0052] (f) Select input and output variables;

[0053] (g) Sample normalization processing;

[0054] (h) RNN network model selection;

[0055] (i) RNN network training;

[0056] (j) RNN network estimation.

[0057] The beneficial effects brought by adopting the above technical solution are:

[0058] The present invention uses a new RNN neural network model based on the existing technology for estimating battery SOP, adds multi-parameter constraints of SOC and battery model to make a new estimate of SOP, and greatly improves the accuracy of the estimation model. Based on the present invention, the utilization rate of battery energy can be effectively improved and the battery life can be extended. It effectively evaluates the key indicators of electric vehicle acceleration and climbing performance, characterizes the limit capacity of power battery charging and discharging power, and is crucial for the maximum energy recovery function of the motor. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is a flow chart of the method of the present invention;

[0060] Figure 2 This is a schematic diagram of an equivalent circuit model of a lithium battery in the present invention;

[0061] Figure 3 This is a lithium battery OCV-SOC calibration curve diagram in the embodiment;

[0062] Figure 4 This is a lithium battery SOC estimation diagram in the embodiment;

[0063] Figure 5 The peak power estimation result of lithium battery discharge in the embodiment;

[0064] Figure 6 The peak power estimation result of lithium battery charging in the embodiment; DETAILED DESCRIPTION

[0065] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings.

[0066] The present invention designs a lithium battery SOP online estimation method based on RNN neural network and multi-parameter constraints, such as Figure 1 As shown, the steps are as follows:

[0067] Step 1, establish a lithium battery equivalent circuit model and dynamic system equation, perform lithium battery OCV test, record lithium battery data under UDDS working conditions, and perform online identification of various parameters of the lithium battery through an extended Kalman filter algorithm; the parameters include internal resistance R0, polarization resistance and polarization capacitance;

[0068] Step 2: Establish an EKF discrete nonlinear system equation based on the lithium battery equivalent circuit model, establish the state equation and observation equation of the lithium battery through the extended Kalman filter, and complete the online estimation of SOC under the cycle condition;

[0069] Step 3: Use a method based on SOC and a method based on a battery model to estimate the continuous peak current, and a method based on SOC and battery model multi-parameter constraints to estimate the continuous peak power;

[0070] Step 4: Using the terminal voltage, current, and SOC under the battery operating conditions as input and the peak power SOP as output, import the RNN neural network model, establish the RNN neural network, and complete the estimation of the peak power of the lithium battery.

[0071] In this embodiment, in the above step 1, the lithium battery equivalent circuit model adopts a second-order RC circuit model, such as Figure 2 As shown, the second-order RC circuit model includes electrochemical polarization internal resistance, electrochemical polarization capacitance, concentration polarization resistance and concentration polarization capacitance. oc Indicates the open circuit voltage of the battery when it is not working, U o It represents the terminal voltage of the battery under working condition; I represents the discharge current; R0 represents the internal ohmic internal resistance of the battery; R1, R2, C1, and C2 are two groups of polarization parameters, which represent the electrochemical polarization internal resistance, the concentration polarization internal resistance, the electrochemical polarization capacitance, and the concentration polarization capacitance, respectively. The two groups of parameters can be used to characterize the fast and slow response characteristics of the battery's internal reaction process, respectively. U1 and U2 represent two groups of RC loop terminal voltages, respectively, and their values ​​and the total voltage of the battery.

[0072] In this embodiment, in the above step 1, if the dynamic system equation of the battery is to be established, it is necessary to count with the SOC of the ampere-hour integration method as a reference, and the corresponding battery second-order model state equation can be described as:

[0073]

[0074]

[0075]

[0076] Among them, SOC(0) is the battery SOC value in the initial state, and SOC(t) is the battery SOC value at time t; is the battery capacity at a specific discharge rate; η is the battery Coulomb coefficient; I(t) is the current at time t; U1(t), U2(t) are the terminal voltages of the two RC combinations at time t; R1, C1, R2, C2 are four polarization parameters.

[0077] The battery dynamic system equation can be described as:

[0078] U0=U oc (t)-R0·I(t)-U1(t)-U2(t)

[0079] Among them, U oc (t) is the open circuit voltage at time t, R0 is the expression of the ohmic internal resistance. I(t) represents the discharge current at time t, U1(t) and U2(t) represent the voltages of the two RC loop terminals at time t.

[0080] In this embodiment, in the above step 1, an 18650 lithium battery is used to perform an OCV test at room temperature, and the test data is fitted to obtain a corresponding relationship curve between the open circuit voltage and the SOC. The specific process is as follows:

[0081] 1a. First, let the lithium battery stand for a long time, and then discharge it at a constant current of 0.3A. When the battery reaches the cut-off voltage (2.8V), the SOC is considered to be 0. After standing for 2 hours, charge it at a constant current and constant voltage of 0.3A and a cut-off current of 0.03A. When the battery reaches the cut-off voltage (4.2V), the SOC is considered to be 100% when the charging current is less than or equal to 0.03A. Stand for 2 hours. Then start the OCV test, first stand for 10s, then discharge it at a constant current for 2h, and then stand for 2h. At this time, the battery SOC is considered to be reduced to 90%. Cycle ten times until the battery SOC is 0, and the OCV test is completed.

[0082] 1b. Perform curve fitting on the open circuit voltage value corresponding to the moment when the SOC value changes from 100% to 10%, and obtain a curve diagram of the relationship between the open circuit voltage and the SOC, such as Figure 3 shown.

[0083] In this embodiment, the specific process of step 1 is as follows:

[0084] 1a. Considering the lithium battery as a system, the state equation in the system can be expressed as:

[0085] x k+1 =f(x k ,u k )+w k y k =g(x k ,u k )+v k

[0086] In the formula, x k is the state variable; u k is the incentive of the system; w k , v k is the system noise; y k is the output of the system; f(x k ,u k ) and g(x k ,u k ) is a nonlinear function.

[0087] 1b. Transform the equivalent circuit state equation into: U O =U OC -U1-U2-R0I.

[0088] 1c. If the relative change of the battery open circuit voltage is zero in a very short time, the derivative of the battery output voltage with respect to time can be obtained:

[0089]

[0090] Among them, U o is the terminal voltage value, t is the time, R1 and C1, R2 and C2 are two sets of electrochemical polarization internal resistance and concentration polarization capacitor, R0 is the internal resistance, U1 is the voltage on a set of polarization internal resistance and concentration polarization capacitor, I is the current value, U oc is the open circuit voltage.

[0091] 1d. Convert the above formula into a state equation, then the system state variable of the battery equivalent circuit is:

[0092] x=[U1 U2 1 / R1 1 / R2 1 / C1 1 / C2]

[0093] Among them, the battery input is: k =I, where u k is the battery terminal voltage at time k.

[0094] 1e. According to the corresponding relationship between the input current and output voltage response curves of the lithium battery, the above formula can be used to obtain the relationship between the state variables of the battery system and the unknown parameters in the system to obtain the parameters under different SOCs.

[0095] 1f. Complete the UDDS working condition and obtain the online parameter identification results.

[0096] In this embodiment, the specific process of step 2 is as follows:

[0097] 2a. Establish the EKF discrete nonlinear system equation:

[0098] χ k+1 =f(x k ,u k )+w k

[0099] y k =g(x l ,u k )+v k

[0100] Among them, χ l+1 is the input, y k is the output, w k and v k is the noise of the system and follows a normal distribution, which is similar to f(x k ,u k ), g(x k ,u k ) are independent of each other, f(xk ,u k ) and g(x k ,u k ) is a nonlinear function.

[0101] 2b. Establish the state equation and observation equation of lithium battery through the extended Kalman filter algorithm:

[0102]

[0103] U k =U oc (s k )-i k R0-U 1,k -U 2,k

[0104] Among them, τ d and τ e are two different time constants, U oc is the open circuit voltage, s k is the estimated SOC value at time k, i k is the current at time k, T is the sampling time, C N is the rated capacity of the battery, η t is the charge and discharge efficiency, U k is the total estimated value of the terminal voltage at time k, U 1,k is the estimated value of the terminal voltage on R1 at time k, U 2,k is the estimated value of the terminal voltage on R2 at time k;

[0105] 2c. Based on the terminal voltage and current data collected during the charging and discharging process of the lithium battery, the discretized state equation and observation equation are subjected to EKF recursive calculation to complete the online estimation of SOC under cyclic conditions.

[0106] First, initialize the system:

[0107]

[0108] The current value I, the current SOC value, and the current terminal voltage value U0 are used as inputs and substituted into the extended Kalman filter algorithm. The input is the battery rated capacity C. N ; Input sampling time T and total step size N.

[0109] Compute the error covariance matrix:

[0110]

[0111] Next, calculate the state prediction:

[0112]

[0113] Update Kalman filter gains:

[0114]

[0115] Corrected state prediction value:

[0116]

[0117] Modified covariance matrix:

[0118] P k / k =(IH k C k ) k / k-1

[0119] At this time, the current value and voltage value at the next moment are obtained, and the SOC is estimated online.

[0120] In this embodiment, the specific process of step 3 is as follows:

[0121] 3a. Calculate the maximum current value that the battery can withstand under the current conditions according to the SOC limit, and multiply it by the corresponding battery terminal voltage to calculate the maximum charge and discharge power. Assume that the battery is discharged at a constant current within a unit time of Δt, and the current value is i k , then the battery SOC at time t+Δt can be expressed as:

[0122]

[0123] Where η i is the battery coulombic efficiency, Q v The actual capacity of the battery. When the battery is working, the SOC value must be within the following range at any time: SOC min <SOC(t)<SOC max , then the maximum charge and discharge current under SOC constraint can be deduced from the above formula:

[0124]

[0125] η chg and η dis is the battery charge and discharge efficiency, is the minimum charging current, is the maximum discharge current. SOC(t) is the SOC value at time t, such as Figure 4 shown.

[0126] From the above derivation process, it can be seen that the method based on battery SOC limitation can calculate the continuous peak power of the battery within a unit time Δt. In essence, it is to ensure the safety of the power battery, so the estimated value has a certain margin. Therefore, the peak current calculated only with the battery SOC as the constraint condition will also make the peak power estimation result too large.

[0127] 3b. Peak current estimation based on second-order RC model

[0128] According to the terminal voltage range: U t,min t t,max , the maximum charge and discharge current can be estimated as:

[0129]

[0130] In the formula, and are the estimates of the peak discharge and charge currents at time k based on the second-order RC model of the battery. L is the continuous sampling time. g(SOC) represents the open circuit voltage of the battery, SOC k+1 Indicates the battery SOC and current i at time k+1 k+1 Functional relationship of Q v Indicates the actual battery capacity. It means the first-order partial derivative of g(SOC) with respect to SOC. 1,k is the estimated value of the terminal voltage on R1 at time k, U 2,k is the estimated value of the terminal voltage on R2 at time k.

[0131] 3c. Continuous peak power estimation under multi-parameter constraints based on SOC and battery model, such as Figure 5 and Figure 6 :

[0132] Find the dynamic peak current that satisfies the dynamic voltage characteristics, current limit, and SOC constraints of the lithium battery:

[0133]

[0134] Among them, i max 、i min are the maximum discharge current and minimum charge current under the assumed conditions, They are the maximum continuous discharge current and minimum continuous charge current under the constraints of the battery model. and They are respectively the estimates of the continuous maximum charge and discharge current obtained under the battery SOC constraint within the current duration (the maximum value of the maximum discharge current and the minimum value of the minimum charge current based on the battery SOC within the continuous sampling time L).​​

[0135] Then estimate the peak power:

[0136]

[0137] Among them, C N is the rated capacity of the battery, L is the continuous sampling time, j represents the initial moment in the continuous sampling time, τ is the time constant, and Δt is the unit time. g(SOC,C N ) represents the open circuit voltage of the battery based on the rated capacity of the battery, SOC(k) is the SOC at time k, represents g(SOC,C N ) calculates the first-order partial derivative of SOC. 1,k is the estimated value of the terminal voltage on R1 at time k, U 2,k is the estimated value of the terminal voltage on R2 at time k. is the maximum discharge peak power and minimum charging peak power under the current constraints.

[0138] In this embodiment, the specific process of step 4 is as follows:

[0139] 4a. Select input and output variables

[0140] The accuracy and computational complexity of the neural network estimation model are closely related to the selection of input variables and the number of variables. In order to improve the estimation accuracy of the RNN network, factors that have a greater impact on the maximum charge and discharge power of lithium-ion batteries should be selected as much as possible, but at the same time, the scale of the network and the requirements of computational complexity should also be taken into account. This paper selects the battery current, terminal voltage, and state of charge SOC under UDDS conditions as the input variables of the RNN network peak power estimation model.

[0141] 4b. Sample normalization

[0142] After selecting the input variables, it is necessary to obtain the training samples required by the RNN network model and normalize these sample data. This paper collects real-time battery input data and selects 2,000 groups of samples, of which 1,900 groups are randomly selected as network training sample data, and the remaining 100 groups of data are used as network estimation verification data.

[0143] 4c. RNN network model selection

[0144] Since the RNN network model for power estimation built in this paper has 4 input variables, the number of network input layer nodes is set to 4, the number of hidden layer nodes is set to 5, and the output layer node is set to 1. The transfer function from the input layer to the hidden layer is selected as the purelin linear function, and the transfer function from the hidden layer to the output layer is selected as the tansig function. The network learning algorithm is selected as trainlm or Levenberg-Marquardt algorithm; the RNN learning rule is selected as learngdm for updating the network weights and thresholds.

[0145] 4d.RNN network training

[0146] The normalized training sample data is imported into the input layer of the RNN network, and then the network is trained. The connection weights between neurons in each layer and the network threshold are set by default, and the network learning rate, maximum number of iterations, accuracy and other parameters can be initialized according to the following formula:

[0147] net.trainParam.show=50

[0148] net.trainParam.lr=0.09

[0149] net.trainParam.epochs=1000

[0150] net.trainParam.goal=0.000001

[0151] net.trainParam.min_grad=le-12

[0152] Among them, net.trianParam.show means that the network displays the training results every 50 trainings; net.trainParam.lr means that the network learning rate is 0.09; net.trainParam.epochs means that the maximum number of iterations of the network loop is 1000; net.trainParam.epochs means that the sum of squares of the error between the network estimated output and the expected output is 0.000001; and training is stopped when the total error reaches 0.000004; net.trainParam.min_grad means that the minimum gradient value of the performance indicator function is 10 -12 . Then the network is trained.

[0153] 4e.RNN network estimation

[0154] Estimate the normalization of sample data:

[0155] inputn_test=map minmax(′apply′,input_test,inputps)

[0156] Use the trained RNN network model to test the estimated samples:

[0157] an=sim(net,inputn_test)

[0158] The network estimation output is denormalized:

[0159] Bpoutput=mapminmax(′reverse′,an,outputps)

[0160] According to the above steps, the peak power of lithium batteries is estimated in the MATLAB environment. The results are as follows: The charging and discharging peak power estimation results of lithium batteries are obtained. It can be seen that the estimation errors of charging and discharging peak power are all within 2%, the average error of charging peak power is 1.76%, and the average error of discharging peak power is 1.69%. It can be seen that the use of RNN neural network model to estimate lithium battery SOP is very accurate.

[0161] The above description is only a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technology can understand and think of any changes or substitutions within the technical scope disclosed by the present invention, which should be included in the scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A lithium battery SOP online estimation method based on RNN neural network and multi-parameter constraints, characterized in that: The following steps are involved: (1) Establishing a lithium battery equivalent circuit model and dynamic system equation, performing lithium battery OCV testing, recording lithium battery data under UDDS conditions, and performing online identification of various parameters of the lithium battery using an extended Kalman filter algorithm; the parameters include internal resistance R0, polarization resistance, and polarization capacitance; (2) The EKF discrete nonlinear system equation is established based on the equivalent circuit model of the lithium battery. The state equation and observation equation of the lithium battery are established through the extended Kalman filter to complete the online estimation of SOC under the cycle condition. (3) Estimate continuous peak current and continuous peak power; (4) Taking the terminal voltage, current and SOC under the battery working condition as input and the peak power SOP as output, the RNN neural network model is imported and the RNN neural network is established to complete the estimation of the peak power of the lithium battery; The specific process of step (2) is as follows: (21) Establish the EKF discrete nonlinear system equation: χ k+1 =f(x k ,u k )+w k y k =g(x k ,u k )+v k Among them, χ k+1 is the input, y k is the output, w k and v k is the noise of the system and follows a normal distribution, which is similar to f(x k ,u k ), g(x k ,u k ) are independent of each other, f(x k ,u k ) and g(x k ,u k ) is a nonlinear function; (22) The state equation and observation equation of lithium battery are established by the extended Kalman filter algorithm: IN k =U oc (with k )-and k R0-U 1,k -IN 2,k Among them, R1 is the electrochemical polarization internal resistance, R2 is the concentration polarization resistance, R0 is the ohmic internal resistance expression, τ d and τ e are two different time constants, U oc is the open circuit voltage, s k is the estimated SOC value at time k, i k is the current at time k, T is the sampling time, C N is the rated capacity of the battery, η t is the charge and discharge efficiency, U k is the total estimated value of the terminal voltage at time k, U 1,k is the estimated value of the terminal voltage on R1 at time k, U 2,k is the estimated value of the terminal voltage on R2 at time k; (23) Based on the terminal voltage and current data collected during the charging and discharging process of the lithium battery, the discretized state equation and observation equation are subjected to EKF recursive calculation to complete the online estimation of SOC under the cycle condition; The specific process of step (3) is as follows: (31) Estimate the peak current based on the battery SOC: Assume that the battery has a current i per unit time Δt. k Discharging, the battery SOC at time t+Δt is as follows: When the battery is working, the battery SOC value at any time satisfies the following constraints: SOC min <SOC(t)<SOC max , then the maximum charge and discharge current under SOC constraint can be deduced from the above formula: Where η i is the battery coulombic efficiency, Q v is the actual capacity of the battery, SOC(t) is the SOC value at time t, SOC min , SOC max are the minimum and maximum values ​​of SOC, η chg and η dis For battery charging and discharging efficiency, is the minimum charging current based on the battery SOC, is the maximum discharge current based on the battery SOC; (32) Based on the second-order RC model of the battery, estimate the peak current: According to the terminal voltage constraint: U t,min t t,max , the maximum charge and discharge current estimate is:​​ In the formula, and are the estimates of the peak discharge and charge currents based on the second-order RC model of the battery at time k, L is the continuous sampling time; g(SOC) represents the open circuit voltage of the battery, SOC k+1 Indicates the battery SOC and current i at time k+1 k+1 Functional relationship of It means the first-order partial derivative of g(SOC) with respect to SOC; (33) Estimate the peak power of lithium battery based on the multi-parameter constraints of battery SOC and battery second-order RC model: First, estimate the peak current under multi-parameter constraints: Among them, i max 、i min are the preset maximum discharge current and minimum charge current, and are respectively the maximum value of the maximum discharge current and the minimum value of the minimum charge current based on the battery SOC within the continuous sampling time L; Then estimate the peak power: Among them, C N is the rated capacity of the battery, j represents the initial moment in the continuous sampling time, τ is the time constant, is the maximum discharge peak power and minimum charge peak power under multi-parameter constraints, g(SOC,C N ) represents the open circuit voltage of the battery based on the rated capacity of the battery, represents g(SOC,C N )Calculate the first-order partial derivative of SOC.

2. According to the lithium battery SOP online estimation method based on RNN neural network and multi-parameter constraints according to claim 1, it is characterized in that, In step (1), the lithium battery equivalent circuit model is a second-order RC circuit model, and the second-order RC circuit model includes an electrochemical polarization internal resistance R1, an electrochemical polarization capacitor C1, a concentration polarization resistor R2, and a concentration polarization capacitor C2; The dynamic system equation is as follows: U0=U oc (t)-R0·I(t)-U1(t)-U2(t) Among them, U0 represents the terminal voltage of the battery in working state, U oc (t) is the expression of open circuit voltage, R0 is the expression of ohmic internal resistance, I is the discharge current, U1 and U2 are respectively represented by two sets of RC loop terminal voltages, and t is time.

3. According to claim 2, the lithium battery SOP online estimation method based on RNN neural network and multi-parameter constraints is characterized in that, In step (1), an OCV test is performed on a 18650 lithium-ion battery at room temperature, and the test data is fitted to obtain a curve of the corresponding relationship between the open circuit voltage and the SOC.

4. According to the lithium battery SOP online estimation method based on RNN neural network and multi-parameter constraints according to claim 2, it is characterized in that, The specific process of step (1) is as follows: (11) Complete the UDDS operating condition test of the lithium battery at room temperature and record the voltage, current and SOC true value; (12) Storing intermediate process estimates; (13) Initialize the parameters and convert the estimated values ​​of the parameters into R0, R1, R2, C1, C2; update the values ​​at the next moment; (14) Calculate the Kalman filter gain matrix and collect the voltage and current data at the next moment; (15) Repeat (12) to (14) to identify the lithium battery model parameters online.

5. According to the lithium battery SOP online estimation method based on RNN neural network and multi-parameter constraints according to claim 1, it is characterized in that, The specific process of establishing the RNN neural network in step (4) is as follows: (a) Select input and output variables; (b) Sample normalization processing; (c) RNN network model selection; (d) RNN network training; (e) RNN network estimation.

Images