CN110346734A - A kind of lithium-ion-power cell health status evaluation method based on machine learning - Google Patents

Abstract

The invention discloses a method for estimating the state of health of a lithium-ion power battery based on machine learning, which is used for estimating the state of charge and the state of health of the power battery in real time. By establishing the equivalent circuit model of the lithium-ion battery, its parameters are identified, and then the Uoc‑SOC model is established to estimate the SOC. Using a large amount of offline data training to obtain a neural network model that takes Uoc-SOC model parameters as input and the maximum available capacity as output. Carry out curve fitting on Uoc and SOC at the same time, obtain the parameters to be identified in the model, input them into the trained neural network model, obtain the maximum available capacity, and return the obtained Uoc‑SOC model parameters and maximum available capacity to Go to the SOC estimation step to update the parameters of its state equation and observation equation. The invention proposes a method for estimating the state of health of a lithium-ion battery, which performs online estimation of the state of health of the battery, and updates parameters for SOC estimation, thereby improving the estimation accuracy.

Description

A method for estimating the state of health of lithium-ion power batteries based on machine learning

technical field

The invention relates to the technical field of state estimation of a power battery management system of an electric vehicle, in particular to joint estimation of a state of charge and a state of health of a power battery.

Background technique

With the depletion of global oil resources and the increasingly serious environmental pollution, electric vehicles, as an energy-saving, environmentally friendly and sustainable transportation tool, have attracted people's attention. As the power source of electric vehicles, the performance of power battery packs has always been the focus of research. When the battery of an electric vehicle is used, it needs to work within a reasonable range of voltage, current, and temperature. Therefore, there is a need to effectively manage the use of electric batteries on electric vehicles. The specific equipment for battery management in electric vehicles is the Battery Management System (BMS). It not only ensures the safe and reliable use of the battery, but also enables the full capacity of the battery and prolongs its life. As a bridge for communication between the battery vehicle controller and the driver, it controls the charging and discharging of the battery pack and provides information to the vehicle control. The device reports the basic parameters and fault information of the power battery system. The level of the battery management system largely determines the performance of the power battery pack. Therefore, a real-time and efficient battery management system is very important.

The battery management system (BMS) is an important link in the development of new energy vehicle power system assembly and large-scale energy storage system. The battery state of charge (SOC) is used to characterize the remaining power of the battery, that is, the percentage of the remaining power to the rated capacity. The state of charge (SOC) of the battery cannot be obtained directly from the battery itself, but can only be estimated indirectly by measuring the external characteristic parameters of the battery pack (such as voltage, current, etc.). During the use of electric vehicle power batteries, due to the internal complex electrochemical reactions, the battery characteristics show a high degree of nonlinearity, which makes it very difficult to accurately estimate the battery state of charge (SOC). Nonlinear Kalman filters (such as extended Kalman filter, unscented Kalman filter, etc.) for dealing with nonlinear problems are considered to estimate SOC. When using these algorithms, the relationship between SOC and Uoc, as well as the maximum available capacity of the battery will be involved. The current method generally does not consider its changes, and defaults to a fixed value. But in fact, as the battery ages, the relationship between SOC and Uoc will change, and the maximum available capacity of the battery will also change. If these changes are not updated in time, the estimation error of SOC will become larger and larger.

SOH refers to the state of health of the battery, that is, the ratio of the current maximum available capacity to the initial maximum available capacity. As the battery usage time increases, the battery will gradually age, and the internal resistance will increase, and the battery capacity will decay. The reasons for battery capacity fading are complex, involve many factors, and change slowly. There is currently no accurate physical model of the recession. When using machine learning to build a model, the choice of health factors has a greater impact on its final accuracy, so it is very important to choose an appropriate health factor.

Contents of the invention

In order to solve the problems existing in the prior art, the present invention jointly estimates the state of charge (SOC) and the state of health (SOH) of the battery, and improves the accuracy of SOC estimation by updating the coefficient and the maximum possible capacity of the relationship between Uoc and SOC in real time , to reduce the estimation error; and the coefficient of the relationship between Uoc and SOC is used as the health factor and the input of the BP neural network model. The invention has great significance for estimating and controlling the state of the whole battery, improving the service life of the battery and fully exerting the capacity of the battery.

The present invention adopts the following technical solutions to achieve the above-mentioned technical purpose.

A method for estimating the state of health of a lithium-ion power battery based on machine learning, comprising the following steps:

Step (1), establishing an equivalent circuit model of a lithium-ion power battery, and identifying unknown parameters in the model;

Step (2), establish Uoc-SOC model: Among them, a _i , _{bi , c i .} are the parameters to be identified in the model, U _OC is the open circuit voltage of the battery, and SOC is the state of charge of the battery;

Step (3), estimating the state of charge SOC;

In step (4), the off-line data is obtained by curve fitting to obtain the values of a _i , b _i , and c _i ;

Step (5), normalize a _i , b _i , _ci and the maximum available capacity C, get a' _i , b' _i , c' _i as input and C' as output, and use machine learning algorithm to a' _i , b' _i , c' _i and the corresponding C' are trained, and finally a machine learning model with a' _i , b' _i , c' _i as input and C' as output is obtained;

In step (6), curve fitting is performed on Uoc and SOC at the same time to obtain input values a _i , _{bi , and c i ,} which are normalized and input to the machine model to obtain the maximum available capacity C;

Step (7), return the values of a _i , b _i , c _i and C obtained in step (6) to step (3), and update the corresponding parameters in the state equation and observation equation.

Further, the equivalent circuit model is selected from a Thevenin equivalent circuit model or a second-order RC equivalent circuit model.

Further, the Uoc-SOC model uses U _OC =K ₀ +K ₁ z+K ₂ z ² +K ₃ z ³ +K ₄ z ⁴ +K ₅ z ⁵ +K ₆ z ⁶ or or or Replacement, where z is the state of charge SOC, K ₀ , K ₁ , K ₂ K ₃ , K ₄ , K ₅ , K ₆ , α ₁ , α ₂ are the parameters to be identified in the model.

Further, the state of charge SOC is estimated using the extended Kalman filter algorithm, and the state equation of the polarization voltage of the battery and the state of charge is Where U _p , R _p , and C _p represent battery polarization voltage, resistance, and capacitance, η is Coulombic efficiency, ΔT is sampling time interval, _IL is battery charge and discharge current, ω _k is system noise; the observation equation of terminal voltage is Where υ _k is the measurement noise.

The beneficial effects of the present invention are: the method for estimating the state of health of lithium-ion power batteries based on machine learning proposed by the present invention can update in real time the parameters in the Uoc-SOC model and the maximum available capacity of the current battery when using the model method to estimate the state of charge, Improve the estimation accuracy of SOC. In addition, the present invention uses the parameters in the Uoc-SOC model as the health factor of the data-driven health state estimation method, which has higher estimation accuracy; At the same time, the updated parameters are used as the input of the SOH estimation, and the SOC estimation and the SOH estimation share the same set of parameters, which reduces the calculation amount.

Description of drawings

Fig. 1 is the equivalent circuit diagram of lithium-ion power battery model;

Fig. 2 is a flow chart of the method for estimating the state of health of a lithium-ion power battery based on machine learning in the present invention.

Detailed ways

The specific technical solutions of the present invention will be further described below in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited thereto.

A method for estimating the state of health of a lithium-ion power battery based on machine learning, comprising the following steps:

Step (1), set up the equivalent circuit model of lithium-ion power battery, can select Thevenin equivalent circuit model for use, or second-order RC equivalent circuit model, the present embodiment takes Thevenin equivalent circuit as example (as shown in Figure 1), wherein , U _OC represents the open circuit voltage of the battery, U _t represents the terminal voltage of the battery, R ₀ represents the ohmic internal resistance of the battery, U _p , R _p , and C _p represent the polarization voltage, resistance, and capacitance of the battery; _IL represents the charge and discharge of the battery current. According to the circuit schematic diagram in Figure 1, the above equivalent circuit model is analyzed by electrotechnical theory, and the continuous time equation of the battery model is established:

U _t =U _oc -I _L R ₀ -U _p (1)

The transfer function is obtained by Laplace transform:

Bilinear transformation, let have to

make

Formula (4) can be simplified as:

U _{L, k} = a ₁ U _{L, k-1} + U _{OC, k} - _{a 1} U _{OC, k-1} + a ₂ I _{L, k} + a ₃ I _{L, k-1} (5)

And because the sampling time T is very short, then:

ΔU _{OC, k} = U _{OC, k} - U _{OC, k-1} ≈ 0 (6)

Then formula (6) can be simplified as:

U _{t, k} = (1-a ₁ ) U _{OC, k} + a ₁ U _{t, k-1} + a ₂ I _{L, k} + a ₃ I _{L, k-1} (7)

Then use the recursive least squares method for parameter identification, formula (7) can be written as:

Among them, Φ _{Ls, k} is the data matrix of the system, and θ _{Ls, k} is the parameter matrix.

Step (2), establish Uoc-SOC model:

Using a battery test cabinet, a 18650-type lithium-ion battery is tested for cycle life, and the values of SOC and Uoc corresponding to different cycle times are obtained. Curve fitting is performed on the obtained data to obtain the relationship between Uoc-SOC: where a _i , b _i and _ci are the parameters to be identified in the model. These parameters to be identified change with the change of the battery health state, and can be used as health factors to characterize the health state.

The Uoc-SOC model can also adopt the following models:

U _OC ＝K ₀ +K ₁ z+K ₂ z ² +K ₃ z ³ +K ₄ z ⁴ +K ₅ z ⁵ +K ₆ z ⁶ (11)

Where z is the state of charge SOC, K ₀ , K ₁ , K ₂ K ₃ , K ₄ , K ₅ , K ₆ , α ₁ , and α ₂ are the parameters to be identified in the model. It can also be used as a health factor that characterizes the state of health.

Take formula (9) as an example below, take n=3, fit it, and obtain the parameters under different cycle times as follows:

Cycles a₁ b₁ c₁ a₂ b₂ c₂ a₃ b₃ c₃ 1 3.936 1.242 0.8202 0.095 0.5856 0.1909 3.004 -0.2176 0.7787 25 2.985 1.196 0.3515 2.863 0.7151 0.4342 3.12 0.04232 0.5155 50 3.51 1.196 0.4159 2.654 0.6447 0.4176 3.061 0.02453 0.482 75 3.471 1.219 0.6137 0.1646 0.6018 0.2187 3.353 0.1179 0.8297

It can be seen that under different cycle times, the parameters to be identified change, which can be used as health factors.

Step (3), estimating the SOC of the state of charge, the present embodiment takes the extended Kalman filter algorithm as an example to estimate the SOC, the polarization voltage of the lithium-ion battery and the state equation of the state of charge such as formula (10), terminal voltage The observation equation of is as formula (11).

Among them: ω _k is the system noise, assuming it conforms to Gaussian distribution, υ _k is the measurement noise, also assuming it conforms to Gaussian distribution, η is Coulomb efficiency, and ΔT is the sampling time interval.

Step (4), obtain a large amount of offline data: test the Uoc and SOC data when the battery is discharged at different discharge rates (such as 0.1C, 0.5C, 1C, 2C) at different temperatures and in different health states, and in various dynamic Uoc and SOC data under working conditions (such as the new standard European cycle test-NEDC, urban road cycle-UDDS); select the Uoc-SOC model of formula (9), and take n=3 in consideration of model accuracy and computational complexity, Then, the values of a _i , b _i , and _ci are obtained by means of curve fitting.

Step (5), since the values of a _{i , b i , ci are different under different health states (that is, the current maximum available capacity C is different), through correlation analysis, select a i , b i ,} The value of ci is taken as input, and C is taken as output; in order to improve the training accuracy, the maximum and minimum normalization method is used to normalize the input a _i , b _i , _ci and the output C to obtain a' _{i ,} b ' _i , c' _i and C'; use neural network (take BP neural network as an example) algorithm to learn a' _i , b' _i , c' _i , C' (i=1, 2, 3;) Training, finally get a BP neural network model with a' _i , b' _i , c' _i as input and C' as output.

Step (6), through steps (1) and (3), real-time Uoc and SOC data at the same time can be obtained, select SOC and corresponding Uoc in the range of 10%-90%, take SOC as the abscissa, and Uoc as the ordinate , the input values a _i , b _i , c _i are obtained by curve fitting, after evolutionary normalization of a _i , b _i , c _i , the BP neural network model obtained in step (5) is input, and the output C' Perform denormalization to obtain the current maximum available capacity C of the lithium battery.

Step (7), return the values of a _i , b _i , c _i and C obtained in step (6) to step (3), and update the corresponding parameters in the state equation and observation equation.

Step (8), step (7) After updating the parameters, execute steps (1), (3), (5), (6), and (7) to estimate the state of health of the battery for the next cycle.

It should be noted that although the content disclosed in the present invention has been described through the above-mentioned embodiments, the described content should not be regarded as limiting the present invention. The modifications and equivalent replacements made by those skilled in the art to the technical solutions of the present invention shall be covered by the scope of the claims of the present invention.

Claims (4)

1. a kind of lithium-ion-power cell health status evaluation method based on machine learning, which is characterized in that including following step It is rapid:

Step (1), establishes the equivalent-circuit model of lithium-ion-power cell, and recognizes the unknown parameter in model；

Step (2), establishes Uoc-SOC model:Wherein a_i、b_i、c_iIt is model In parameter to be identified, U_OCIndicate the open-circuit voltage of battery, SOC is battery charge state；

Step (3), estimates state-of-charge SOC；

Step (4) obtains a by off-line data by curve matching_i、b_i、c_iValue；

Step (5), to a_i、b_i、c_iAnd maximum available C is normalized, and obtains a '_i、b′_i、c′_iAs input, C' As output, using machine learning algorithm to a '_i、b′_i、c′_iAnd corresponding C' is trained, and is finally obtained with a '_i、b′_i、c′_i To input the machine learning model for C' being output；

Step (6) carries out curve fitting to obtain input value a to the Uoc and SOC of synchronization_i、b_i、c_i, it is input to after normalization Machine mould obtains maximum available C；

Step (7), by a obtained in step (6)_i、b_i、c_iAnd C value returns to step (3), updates state equation and observational equation In corresponding parameter.

2. a kind of lithium-ion-power cell health status evaluation method based on machine learning according to claim 1, It is characterized in that, the equivalent-circuit model selects Thevenin equivalent-circuit model or Order RC equivalent-circuit model.

3. a kind of lithium-ion-power cell health status evaluation method based on machine learning according to claim 1, It is characterized in that, the Uoc-SOC model U_OC=K₀+K₁z+K₂z²+K₃z³+K₄z⁴+K₅z⁵+K₆z⁶OrOrOrReplacement, wherein z is state-of-charge SOC, K₀、K₁、K₂ K₃、 K₄、K₅、K₆、α₁、α₂For parameter to be identified in model.

4. a kind of lithium-ion-power cell health status evaluation method based on machine learning according to claim 1, It is characterized in that, the state-of-charge SOC is estimated using expanded Kalman filtration algorithm, the polarizing voltage of battery and charged shape The state equation of state isWherein U_p、R_p、C_p Indicate that battery polarization voltage, resistance, capacitor, η are coulombic efficiency, Δ T is sampling time interval, I_LFor battery charging and discharging stream, ω_k For system noise；End voltage observational equation beWherein υ_k To measure noise.