Research on real-time monitoring technology of battery health state during electric vehicle switching process

This paper compares the global electric vehicle sales and penetration rate from 2013 to 2022 to analyze the development trend of electric vehicles, and the development trend of new energy electric vehicles is shown in Figure 1. From the perspective of the overall sales of new energy vehicles in recent years, the sales of new energy electric vehicles have continued to rise, and there is a continuous upward momentum. In 2019, the total global sales of electric vehicles exceeded 1 million units, and the global penetration rate of electric vehicles reached about 1.5%. By 2022, the sales of new energy electric vehicles surged to 2.35 million units, and the global penetration rate was 2.68%. Although traditional fuel vehicles are still the absolute mainstream, new energy electric vehicles still have a lot of market space, and with the continuous support of the state and the government, the demand for new energy vehicles will continue to increase. Based on this background, this paper explores the power exchange technology of electric vehicles.

Figure 1.

The development trend of new energy electric cars

The electric vehicle power exchange supply mode is shown in Figure 2. According to the category, specification parameters, and characteristics of the battery pack in new energy electric vehicles, there are some differences in the charging mode of these vehicles. Electric vehicles can be charged by DC or AC charging for power replenishment. In the process of electric vehicle power exchange, AC power usually adopts 220V voltage, the current is not big, within 16A, the charging time is usually 6-10 hours, and the charging speed is slow. DC power usually uses 380V voltage, the current is larger, and in 100-200A, the general charging time is 1~2 hours. A short period to replenish a large amount of power is likely to lead to battery heating and the impact on battery life. It is not difficult to find advantages and disadvantages of AC charging and DC charging. The battery life of the AC charger is long, but the charging speed is not fast enough to meet urgent operational needs. DC charging (fast charging) is fast, but it has a high energy consumption, which has an impact on battery life. Furthermore, the current intensity during the charging process is high and easily tangled.

Figure 2.

Power supply model diagram

Parameter identification for the battery model is a key link in modeling, which has a direct impact on the accuracy of the built model. The pulse current method is a commonly used parameter identification method, which is simple in principle and easy to implement, but its testing conditions are too simple, and the established model cannot accurately simulate the characteristics of the battery under the drastic change of the current conditions, and the error is large. The least squares (LS) method is widely used for identification purposes. It matches the best function of the data by minimizing the sum of squared errors. It is simple to compute and has fast convergence speed. Therefore, this paper uses the LS in combination with the designed parameter identification conditions to identify the parameters of the battery model.

RLS-based model parameter identification assumes discrete systems that require parameter identification: (9)A(z−1)Y(k)=B(z−1)U(k)+e(k)$$A({z^{ - 1}})Y(k) = B({z^{ - 1}})U(k) + e(k)$$

Where. (10)A(z−1)=1+a1z−1+a2z−2+...+anz−n$$A({z^{ - 1}}) = 1 + {a_1}{z^{ - 1}} + {a_2}{z^{ - 2}} + ... + {a_n}{z^{ - n}}$$ (11)B(z−1)=1+b1z−1+b2z−2+...+bnz−n$$B({z^{ - 1}}) = 1 + {b_1}{z^{ - 1}} + {b_2}{z^{ - 2}} + ... + {b_n}{z^{ - n}}$$

Y(k) is the system output observation at moment k and U(k) is the system input observation at moment k. Rewrite the model in least squares form: (12)Y(k)=−a1Y(k−1)−a2Y(k−2)−...−anY(k−n)+U(k)+b1a1U(k−1)+...+bnU(k−n)+e(k)$$\begin{array}{rcl} Y(k) = - {a_1}Y(k - 1) - {a_2}Y(k - 2) - ... - {a_n}Y(k - n) \\\quad \quad \quad + U(k) + {b_1}{a_1}U(k - 1) + ... + {b_n}U(k - n) + e(k) \\ \end{array}$$

Let h(k)=[Y(k−1)Y(k−2)…Y(k−n)U(k)U(k−1)…U(k−n)]$$h(k) = \left[ {\begin{array}{*{20}{c}} {Y(k - 1)}&{Y(k - 2)}& \ldots &{Y(k - n)}&{U(k)}&{U(k - 1)}& \ldots &{U(k - n)} \end{array}} \right]$$, θ(k)=[−a1−a2…−an1b1…bn]T$$\theta (k) = {\left[ {\begin{array}{*{20}{c}} { - {a_1}}&{ - {a_2}}& \ldots &{ - {a_n}}&1&{{b_1}}& \ldots &{{b_n}} \end{array}} \right]^{\text{T}}}$$. h(k) be the system data vector and Cθ(k) be the parameter vector to be recognized. Then Eq. can be rewritten as: (13)Y(k)=h(k)θ(k)+e(k)$$Y(k) = h(k)\theta (k) + e(k)$$

Assuming that there are N sets of system input and output observations, the inputs and outputs of the system can be extended to the N dimension: (14)YN(k)=hN(k)θ+eN(k)$${Y_N}(k) = {h_N}(k)\theta + {e_N}(k)$$

Style: (15)YN(k)=[Y(k)Y(k+1)⋮Y(k+N)]$${Y_N}(k) = \left[ {\begin{array}{*{20}{c}} {Y(k)} \\ {Y(k + 1)} \\ \vdots \\ {Y(k + N)} \end{array}} \right]$$ (16)hN(k)=[Y(k−1)...Y(k−n)U(k)...U(k−n)Y(k)...Y(k+1−n)U(k+1)...U(k+1−n)⋮⋮⋮⋮⋮⋮Y(k+N - 1)...Y(k+N0n)U(k+N)...U(k+N−n)]$${h_N}(k) = \left[ {\begin{array}{*{20}{c}} {Y(k - 1)}&{...}&{Y(k - n)}&{U(k)}&{...}&{U(k - n)} \\ {Y(k)}&{...}&{Y(k + 1 - n)}&{U(k + 1)}&{...}&{U(k + 1 - n)} \\ \vdots & \vdots & \vdots & \vdots & \vdots & \vdots \\ {Y(k + N{\text{ - 1}})}&{...}&{Y(k + N{\text{0}}n)}&{U(k + N)}&{...}&{U(k + N - n)} \end{array}} \right]$$ (17)eN(k)=[e(k)e(k+1)⋮e(k+N)]$${e_N}(k) = \left[ {\begin{array}{*{20}{c}} {e(k)} \\ {e(k + 1)} \\ \vdots \\ {e(k + N)} \end{array}} \right]$$

with the sum of squares of the residuals as the criterion function: (18)J(θ)=∑k=1Ne^2(k)$$J(\theta ) = \sum\limits_{k = 1}^N {{{\hat e}^2}} (k)$$

Of these, e^(k)=Y(k)−h(k)θ$$\hat e(k) = Y(k) - h(k)\theta$$.

Least squares estimation is the optimal estimation in the sense of minimizing the criterion function. Rewriting Eq: (19)J(θ)=(YN(k)−hN(k)θ)T(YN(k)−hN(k)θ)$$J(\theta ) = {({Y_N}(k) - {h_N}(k)\theta )^{\text{T}}}({Y_N}(k) - {h_N}(k)\theta )$$

The partial derivative of Eq. The value θ^$$\hat \theta$$ that can make the partial derivative equal to 0, i.e., the value that minimizes J(θ), is the least squares estimate, which is calculated as follows (20)θ^=(hNThN)−1hNTYN$$\hat \theta = {({h_N}^{\text{T}}{h_N})^{ - 1}}{h_N}^{\text{T}}{Y_N}$$

Since solving θ^$$\hat \theta$$ requires an inverse operation on θ^$$\hat \theta$$, hNThN$$h_N^{\text{T}}{h_N}$$ cannot invert h_N when it is not a full rank matrix, resulting in an inability to solve θ^$$\hat \theta$$. With the increase of the system dimension N, the computation of LS will gradually increase and occupy a large amount of system memory. The drawbacks of LS are improved by reducing it to a recursive form of the algorithm, i.e., RLS. RLS corrects the estimation results of the previous step using the system input and output values at the latest moment, which reduces the computational difficulty of the algorithm by eliminating the need to obtain the optimal results in one operation.

The idea of RLS is to derive the form of θ^(k)=θ^(k−1)+$$\hat \theta (k) = \hat \theta (k - 1) +$$ correction from Eq. θ^(k)$$\hat \theta (k)$$ and θ^(k−1)$$\hat \theta (k - 1)$$ are the LS estimates obtained from the first k and the first k − 1 samples, respectively. The specific derivation of RLS is as follows, and let P⁻¹(k) = h_N(k)^Th_N(k), then: (21)P−1(k)=∑i=1kh(i)h(i)T=∑i=1k−1h(i)h(i)T+h(k)h(k)T=P−1(k−1)+h(k)h(k)T$${P^{ - 1}}(k) = \sum\limits_{i = 1}^k h (i)h{(i)^{\text{T}}} = \sum\limits_{i = 1}^{k - 1} h (i)h{(i)^{\text{T}}} + h(k)h{(k)^{\text{T}}} = {P^{ - 1}}(k - 1) + h(k)h{(k)^{\text{T}}}$$

The transformation of hNTYN$${h_N}^T{Y_N}$$ yields (22)hNTYN=∑i=1kh(i)Y(i)=∑i=1k−1h(i)Y(i)+h(k)Y(k)=hN(k−1)TYN(k−1)+h(k)Y(k)$$\begin{array}{rcl} {h_N}^T{Y_N} &=& \sum\limits_{i = 1}^k h (i)Y(i) = \sum\limits_{i = 1}^{k - 1} h (i)Y(i) + h(k)Y(k) \\ &=& {h_N}{\left( {k - 1} \right)^{\text{T}}}{Y_N}\left( {k - 1} \right) + h(k)Y(k) \\ \end{array}$$

The LS estimate at moment k-1 is transformed to: (23)P−1(k−1)θ^(k−1)=hN(k−1)TYN(k−1)$${P^{ - 1}}(k - 1)\hat \theta (k - 1) = {h_N}{(k - 1)^{\text{T}}}{Y_N}(k - 1)$$

Combining Eq. simplifies for AA and gives: (24)θ^(k)=θ^(k−1)+K(k)[Y(k)−h(k)Tθ^(k−1)]$$\hat \theta (k) = \hat \theta (k - 1) + K(k)[Y(k) - h{(k)^{\text{T}}}\hat \theta (k - 1)]$$

Among them: (25)K(k)=P(k−1)h(k)[1+h(k)P(k−1)h(k)T]−1$$K(k) = P(k - 1)h(k){[1 + h(k)P(k - 1)h{(k)^{\text{T}}}]^{ - 1}}$$

Perform the transformation to get: (26)P(k)=P(k−1)[E−K(k)h(k)T]$$P(k) = P(k - 1)[E - K(k)h{(k)^{\text{T}}}]$$

As the number of iterations increases, the RLS will be saturated with data, and the latest observation data will gradually lose its influence on parameter identification, which affects the effect of parameter identification, so the role of historical data can be weakened by introducing the forgetting factor λ in the RLS. The calculation process of RLS after the introduction of the forgetting factor is as follows: (27)K(k)=P(k−1)h(k)[λ+h(k)P(k−1)h(k)T]−1$$K(k) = P(k - 1)h(k){[\lambda + h(k)P(k - 1)h{(k)^{\text{T}}}]^{ - 1}}$$ (28)P(k)=1λP(k−1)[E−K(k)h(k)T]$$P(k) = \frac{1}{\lambda }P(k - 1)[E - K(k)h{(k)^{\text{T}}}]$$ (29)θ^(k)=θ^(k−1)+K(k)[Y(k)−h(k)Tθ^(k−1)]$$\hat \theta (k) = \hat \theta (k - 1) + K(k)[Y(k) - h{(k)^{\text{T}}}\hat \theta (k - 1)]$$

In general, the value of λ should be close to 1. For linear systems, the value of λ should be in the range of [0.95, 1], and the smaller its value is, the faster the algorithm forgets and the more real-time the identification results are, and the larger its value is, the more the historical data affect the identification results, and the algorithm will be degraded to an ordinary RLS when λ = 1.

Before using RLS for parameter identification, the identification parameter matrix θ and covariance matrix P need to be initialized, i.e., θ(0) and P(0).The values of θ(0) can be taken empirically to be in the approximate range of the parameters capable of providing a high algorithmic convergence rate, such that P(0) = α²E, α are sufficiently large real numbers.

Battery SOC estimation is a key function of BMS, and accurate SOC estimation is crucial for the safe and reliable operation of battery systems. Based on the established second-order Thevenin model, the application of EKF in estimating battery SOC is investigated to improve the shortcomings of EKF and enhance the performance of the algorithm in estimating SOC. 1)

Second-order Thevenin equivalent circuit model state space equations

The state-space equations of the system are developed based on the structure of the second-order Thevenin battery model for SOC estimation. The main variables required by the algorithm are first selected. The battery end voltage U and operating current I can be obtained by direct sampling using sensors. Usually, the operating current is selected as the input variable and the end voltage as the output variable.

The terminal voltage can be calculated from the open-circuit voltage U_oc, the polarization voltage U₁,U₂, and the ohmic internal resistance voltage drop IR₀. The open-circuit voltage is a function of soc, so SOC,U₁ and SOC,U₁ are chosen as state variables. According to the structure of the battery model, it can be obtained by Kirchhoff’s law: (30)I=U1R1+C1dU1dt=U2R2+C2dU2dt$$I = \frac{{{U_1}}}{{{R_1}}} + {C_1}\:\frac{{d{U_1}}}{{dt}} = \frac{{{U_2}}}{{{R_2}}} + {C_2}\:\frac{{d{U_2}}}{{dt}}$$ (31)U=f(SOC)−U1−U2−IR0$$U = f(SOC) - {U_1} - {U_2} - I{R_0}$$

The voltage full response of the RC parallel network under dc excitation is: (32)Ui=Ui0e−1τi+IRi(1−e−1τi)i=1,2$${U_i} = {U_{i0}}{e^{ - \frac{1}{{{\tau _i}}}}} + I{R_i}(1 - {e^{ - \frac{1}{{{\tau _i}}}}})\quad i = 1,2$$

where U_i0 is the initial voltage of the RC shunt network and IR_i is the final steady state voltage of the RC shunt network.

The SOC based on the ampere-time integration method is calculated as: (33)SOC=SOC0−1Cn∫0IIdτ$$SOC = SO{C_0} - \frac{1}{{{C_n}}}\int_0^I {Id\tau }$$

The equations of state and observation equations of the battery model are obtained by discretization: (34)x(k)=A(k−1)x(k−1)+B(k−1)I(k−1)+w(k−1)$$x(k) = A(k - 1)x(k - 1) + B(k - 1)I(k - 1) + w(k - 1)$$ (35)U(k)=f(SOC(k))−U1(k)−U2(k)−I(k)R0+ν(k)$$U(k) = f(SOC(k)) - {U_1}(k) - {U_2}(k) - I(k){R_0} + \nu (k)$$

Among them: (36)A(k−1)=[1000e−T/t1000e−T/t2]$$A(k - 1) = \left[ {\begin{array}{*{20}{c}} 1&0&0 \\ 0&{{e^{ - T/{t_1}}}}&0 \\ 0&0&{{e^{ - T/{t_2}}}} \end{array}} \right]$$ (37)B(k−1)=[−T/CnR1(1−e−T/t1)R2(1−e−T/t2)]$$B(k - 1) = \left[ {\begin{array}{*{20}{c}} { - T/{C_n}} \\ {{R_1}(1 - {e^{ - T/{t_1}}})} \\ {{R_2}(1 - {e^{ - T/{t_2}}})} \end{array}} \right]$$ (38)x(k)=[SOC(k)U1(k)U2(k)]$$x(k) = \left[ {\begin{array}{*{20}{c}} {SOC(k)} \\ {{U_1}(k)} \\ {{U_2}(k)} \end{array}} \right]$$

T is the sampling period, τ₁, τ₂ is the time constant, and τ₁ = R₁C₁, τ₂ = R₂C₂.

The open-circuit voltage U_oe and the SOC as a function of f(SOC(k)) in the observation equation are nonlinear. If the EKF algorithm is used to estimate the SOC, it is necessary to carry out a first-order Taylor expansion of the observation equation at the initial estimated value of the state variable, x^(k/k−1)$$\hat x(k/k - 1)$$, to obtain the output matrix, C(k), and then ignore the functions that are not related to the x(k), to obtain the linearized observation equation of the system: (39)C(k)=[∂U∞∂SOC−1−1]|x=x^(k/k−1)$$C(k) = \left[ {\begin{array}{*{20}{c}} {\frac{{\partial {U_\infty }}}{{\partial SOC}}}&{ - 1}&{ - 1} \end{array}} \right]{|_{x = \hat x(k/k - 1)}}$$ (40)U(k)=C(k)x(k)−I(k)R0+ν(k) $$U(k) = C(k)x(k) - I(k){R_0} + \nu (k)$$ 2)

Principle of extended Kalman filter algorithm

The core idea of KF is to estimate the current moment’s value based on the system state space equations and the optimal estimated value of the previous moment’s state, and then use the observable inputs and outputs of the system to correct the current moment’s estimated value to obtain the optimal estimated value of the current moment. The basic requirement of the algorithm is to minimize the quadratic function of the error between the optimal estimates of the state variables of the system and the corresponding real values. KF is a recursive process, which is very suitable for real-time estimation, and therefore the algorithm has been used in many engineering practices. KF is represented by the following state-space equation for a linear system: (41)x(k)=A(k−1)x(k−1)+B(k−1)u(k−1)+w(k−1)$$x(k) = A(k - 1)x(k - 1) + B(k - 1)u(k - 1) + w(k - 1)$$ (42)y(k)=C(k)x(k)+D(k)u(k)+v(k)$$y(k) = C(k)x(k) + D(k)u(k) + v(k)$$

KF obtains x^(k/k−1)$$\hat x(k/k - 1)$$ by preliminary estimation of state variable x, generally there will be a certain estimation error, so the algorithm is based on the correction of the preliminary estimation value to get a more accurate a posteriori estimation value, the amount of correction K(k)[y(k)−y^(k/k−1)]$$K(k)[y(k)-\hat y(k/k-1)]$$ is determined by the Kalman gain K(k) and the output error y(k)−y^(k/k−1)$$y(k)-\hat y(k/k-1)$$, K(k) plays a role in determining the trust weights of the state variable correction value x^(k−1)$$\hat x(k - 1)$$ in the previous moment and the observation value y(k) in the current moment, the bigger JJ is, the bigger the trust weight of the observation value is, and the smaller the trust weight of the state variable is. The larger K(k) is, the larger the trust weight of the observation is, and the smaller the trust weight of the state variable is.KF uses a combination of a priori estimation and a posteriori correction to ensure the accuracy of the estimation results.

The traditional KF can only be applied to linear models, which is often dealt with by converting a nonlinear system into an approximately linear system through the technique of linearization and then carrying out the filtering operation, and the EKF is a typical method on which the KF is based. The core idea of EKF is that for a nonlinear system, the Taylor series is expanded by ignoring the second-order terms and above and retaining only the constant term and the first-order term of the series so as to transform the system into a linear system and then the KF is used to complete the filtering operation.

Assume that the state space equation of a nonlinear system is shown below: (43)x(k)=f[x(k−1),u(k−1)]+w(k−1)$$x(k) = f\left[ {\begin{array}{*{20}{c}} {x(k - 1),u(k - 1)} \end{array}} \right] + w(k - 1)$$ (44)y(k)=h[x(k),u(k)]+v(k)$$y(k) = h[x(k),u(k)] + v(k)$$

where f[x,u] is the nonlinear state function of the system and h[x,u] is the nonlinear observation function of the system. The Taylor series expansion of the nonlinear functions f[x,u] and h[x,u], ignoring the second order and above terms, gives the following form: (45)f[x(k−1),u(k−1)]=f[x^(k−1),u(k−1)]+∂f(x,u)∂x|x=x^(k−1)[x(k−1)−x^(k−1)]$$\begin{array}{l} f\left[ {x(k - 1),u(k - 1)} \right] \\ = f\left[ {\hat x(k - 1),u(k - 1)} \right] + \frac{{\partial f(x,u)}}{{\partial x}}{|_{x = \hat x(k - 1)}}[x(k - 1) - \hat x(k - 1)] \\ \end{array}$$ (46)h[x(k),u(k)]=h[x^(k),u(k)]+∂h(x,u)∂x|x=x^(k)[x(k)−x^(k)]$$h\left[ {x(k),u(k)} \right] = h\left[ {\hat x(k),u(k)} \right] + \frac{{\partial h(x,u)}}{{\partial x}}{|_{x = \hat x(k)}}[x(k) - \hat x(k)]$$

Order. (47)A(k−1)=∂f(x,u)∂x|x=x^(k−1),C(k)=∂h(x,u)∂x|x=x^(k)$$A(k - 1) = \frac{{\partial f(x,u)}}{{\partial x}}{|_{x = \hat x(k - 1)}},C(k) = \frac{{\partial h(x,u)}}{{\partial x}}{|_{x = \hat x(k)}}$$

A(k)and C(k) is the Jacobian matrix, which can be written as: (48)x(k)=A(k−1)x(k−1)+f[x^(k−1),u(k−1)]−A(k−1)x^(k−1)+w(k−1)$$x(k) = A(k - 1)x(k - 1) + f\left[ {\hat x(k - 1),u(k - 1)} \right] - A(k - 1)\hat x(k - 1) + w(k - 1)$$ (49)y(k)=C(k)x(k)+h[x^(k),u(k)]−C(k)x^(k)+v(k)$$y(k) = C(k)x(k) + h[\hat x(k),u(k)] - C(k)\hat x(k) + v(k)$$

After the approximate linearization of the nonlinear system, the recursive procedure that can be obtained by combining the filtering operation of KF according to Eq. is as follows: 1)

Initialize the state variable x and the error covariance matrix P.

Through the analysis of the principle of EKF, EKF estimation of SOC in each filtering cycle needs to derive the nonlinear function and calculate the Jacobian matrix, which increases the number of operations. The algorithm needs to assume that the noise is a known Gaussian white noise, which ignores the change of noise in practical applications, and the robustness of non-Gaussian observation noise is poorer. This paper improves the problems of the EKF and proposes an optimization method based on the Kalman filter algorithm optimization method.

In the actual battery data acquisition process, it is difficult to realize a long-term battery discharge and usually only can be collected for a short period of time within incomplete battery operating data. This section will be a continuous record of the battery operating conditions of the data is called “segment data”. In practice, it is necessary to obtain random segment data by conducting a small (55 sampling points length, i.e., 500s) constant-current discharge experiment on the battery to be predicted, and the discharge current size of the constant-current discharge experiment is the same as that of the constant-current curve used to establish the battery aging database. In this paper, the method proposed in this chapter is validated by forming a sequence of randomized segment data with every 25 sampling points on the constant current discharge curve of the battery as a group and then increasing 20 sampling points sequentially. The random segment data is selected as shown in Fig. 4, and the selected random segment data can be located in any position of the constant current discharge curve cycle, which is more in line with the actual situation.

Figure 4.

Random fragment data selection

When constructing a battery aging database, all the discharge data from the unused state of the battery to the point where it has reached as many cycles as possible are needed, so the battery is subjected to cyclic charging and discharging experiments. Charging once and discharging once is called one charge-discharge cycle. According to the method of this paper and the data provided by the battery, a total of nine constant-current discharge curves are selected to construct the aging database. The SOH of these nine constant-current discharge curves are 92.83%, 90.14%, 87.59%, 85.02%, 79.78%, 74.54%, 69.57%, 67.21%, and 64.68%, respectively, and the nine curves are numbered from 1 to 9 in order, and the constructed The battery aging database is shown in Fig. 5. Each constant current discharge curve corresponds to different battery maximum available capacity, SOH, and other key battery aging state information. When the random segment data matches the closest constant current discharge curve in the aging database, it is considered the complete constant current discharge curve for battery aging information generation. For the same type of battery, in the same degree of aging, its state parameters are independent of the battery operating conditions, so the state parameters of the battery under complex and standard operating conditions are approximately the same, i.e., the method proposed in this paper is also applicable to complex operating conditions.

Figure 5.

Battery status database

In the process of battery discharge, if the current is disconnected and the battery is left standing for a long enough period of time, with the end of capacitor discharge, the terminal voltage gradually rises and is close to the open-circuit voltage, and at this time, the measured terminal voltage value can be taken as the open-circuit voltage value. In this paper, the static method is chosen to measure the SOC-OCV curve, and the experiment is carried out according to the principles of the static method.

In order to verify that the equivalent circuit model and parameter identification results selected in this paper can better reflect the dynamic characteristics of the battery during operation, constant current and pulse current are used for verification, respectively, and the verification results are shown in Fig. 6, where (a) and (b) are the constant-current case and the pulse case, respectively. As can be seen from the figure, the error percentages in both cases are between -0.4% and 0.3%, and the error percentages of the battery model end voltage are controlled within 1%, which is less than 4%. This paper’s selected equivalent circuit model and parameter identification results meet the accuracy requirements and support the application of the EKF algorithm, as shown in the results.

Figure 6.

Constant current condition and pulse condition voltage contrast and error

In this section, the constant-current discharge data of the battery with SOH=88.12% is taken as an example, and 30 sets of segment data with a length of 50 sampling points are intercepted for experiments. All the experiments were conducted at 25°C. The constant current discharge curve with SOH=88.15% corresponds to the actual current maximum usable capacity Q of 1.7623Ah, and the sampling time interval is 9.4s, totaling 339 sampling points. The discharge data were traversed in such a way that every 50 sampling points were a group, and 10 sampling points were added backward each time, totaling 30 groups of random segment data, and the aging database matching results were shown in Fig. 7. It means that 27 out of 30 sets of random segment data are matched to the 3rd aging data curve with SOH of 87.6%. The curve corresponding to the random segment data has an actual SOH of 88.1% and a current maximum usable capacity of 1.76Ah. The matched 3rd aging curve has an SOH of 87.6% and a current maximum usable capacity of 1.75Ah, which matches out to a ∆SOH of 0.54% and a ∆Q of 0.011Ah.

Figure 7.

Aging database matching results

The matching results with the 3rd and 4th curves of the aging database are shown in Fig. 8, and (a) and (b) are the matching results and matching interpolation, respectively. The figure indicates that the remaining 3 sets of random fragment data, i.e., numbered 13, 24, and 26, are matched to the 4th aging data curve, which has a SOH of 85.05% and a maximum usable capacity of 1.7 Ah, and the matched ∆SOH is 3.1% and ∆Q is 0.062 Ah.

Figure 8.

Curve matching

From the previous analysis, it can be seen that since the random segment data is extracted from the curve with a SOH of 88.15% and the maximum available capacity of 1.76Ah, theoretically, it should be more matched with the 3rd aging data curve. The simulation results show that 27 out of 30 groups of random segment data are more compatible with the 3rd aging data curve, so the method proposed in this paper is feasible.

Take the 3rd group of random segment data (sampling points from 20-74) as an example for SOC estimation. From the previous analysis, it can be seen that the best position of the 3rd group of random fragment data matched in the aging database is the 3rd aging curve sampling points from 17-71. At this time, the initial battery’s SOC is 93%. Based on the matched position, SOC estimation is carried out using EKF, and the results of the SOC estimation for the random segment data are shown in Fig. 9. As can be seen from the figure, the predicted value can follow the theoretical value well after a short cycle, and the SOC error is always stabilized within 2%, which illustrates the effectiveness and accuracy of the battery health state monitoring method in this paper.

Figure 9.

Random fragment data soc estimates

The electric vehicle power battery pack online monitoring system is a device designed for real-time online monitoring of electric vehicle power battery packs charging and discharging using sensors, transformers, displays, DSP, and other components. By measuring, analyzing and processing each charging and discharging parameter of the power battery pack, the working status and performance of the battery pack can be displayed in real time, and the SOC capacity of the battery pack can be estimated more accurately in real time by using an improved SOC estimation algorithm, and a more effective charging and discharging equalization system can be designed based on the studied algorithm, which can further improve the operating efficiency of the system and extend the service life of the power battery.

The real-time battery health status monitoring system is shown in Figure 10. The whole device takes the TMS320F28335DSP as the processor core, measures the individual voltage, overall voltage, overall current, and parameters such as internal resistance, insulation, and temperature of the power battery pack with sensors, real-time monitors and analyzes the current operating status of the battery pack, and utilizes an intelligent SOC estimation model transplanted into the DSP to estimate the current SOC capacity of the battery pack based on the overall battery pack voltage, individual voltage, current, and temperature parameters, and reasonably arranges the charging time and location. Using the intelligent SOC estimation model transplanted into the DSP, the current SOC capacity of the battery pack is estimated based on the overall voltage, individual voltage, current, temperature and other parameters of the battery pack, and the charging time and location are rationally arranged. The consistency of the battery pack is analyzed through real-time measured parameters, and the intelligent equalization control strategy configuration model is used to achieve equalization control of the battery pack. Combined with internal resistance and discharge consistency, the analysis of the battery pack’s service time, loss, performance, and other indicators can be conducted. The internal resistance and insulation are obtained by the experimental method, and the parameters do not change much in a short time, so the internal resistance and insulation design system adopts the measured value of input professional equipment as the benchmark, and the measured value of the system is used as the reference.

Figure 10.

System block diagram

The hardware design includes the selection of grid voltage sensors and current sensors, the selection of DSP models, the design of charging measurement circuits, the design of battery pack measurement circuits, and the design of display circuits and communication circuits. The specific hardware circuits include the peripheral circuits of the TMS320F28335 processor, the sampling circuits of the power battery pack, the sampling circuits of the grid parameters and the display circuits. Battery pack measurement using group measurement to 8 batteries as a group, the single voltage through the two MAXIM MAX147528 option 1 analog switches were connected to the positive and negative poles of each battery, the output of the analog switch is connected to a TI INA148UA high common-mode differential operational amplifiers consisting of high common-mode voltage differential amplifier circuit, based on the output of differential circuit Calculated. The overall current is measured using a current transformer as well as a current measurement circuit consisting of MAXIM’s bi-directional, precision current sensing amplifiers MAX471/MAX472, and an absolute value circuit. The overall voltage is measured using a voltage transformer and an AD converter chip. The measurement of the battery’s internal resistance is based on the equalization of discharge and intelligent single charging circuit design, mainly also with MAXIM’s bi-directional, precision current sensing amplifier MAX471/MAX472 composition of the precision measurement circuit.

The system design includes a timer system initialization subroutine, interrupt processing subroutine, sampling subroutine, data processing subroutine, display subroutine and communication subroutine. In order to meet the system’s needs, real-time monitoring of the use of battery packs and charging, as well as monitoring various parameters of the power grid, is necessary. The software design of the system follows a modular design method. Each subroutine runs in different cycles by querying the flag bit, which can not only meet different sampling requirements but also reduce the burden of the whole system while taking into account the real-time and rapidity so that the system can be uninterrupted sampling. The software is written in C. The idea of software design is mainly top-down, modular design, and each sub-module is designed one by one. The integrated development environment CCS provided by TI provides tools for configuring, building, debugging, tracing and analyzing the program, which facilitates the compilation and testing of real-time, embedded signal processing programs. DSP-based application development consists of four basic phases: design, code programming and compilation, debugging, analysis, etc. CCS, the integrated development environment offered by TI, has tools for configuring, building, debugging, tracing, and analyzing programs, which makes it easier to prepare and test real-time, embedded signal processing programs.

In order to verify the effectiveness of the battery detection system in this paper, its practical application in the real-time analysis of the state of the battery and find, through the aging degree of the battery curve test, the estimated battery SOH value shows the current aging level of the battery. Take the battery power as the battery factory rated usable power, and then after completing 30% of the charging of the rated usable power corresponds to the SOH standard of charging at this time, the estimated results and the aging of the state of comparative analysis.

The results of the battery state data are shown in Table 1, from which it can be seen that the monitoring errors under different aging states of the current battery are controlled within 1%, proving that the battery model estimated by SOH data can more effectively deduce the results of battery aging.

The system sensor measurement temperature basically matches. Only when the temperature changes rapidly can there be a difference. Constant current charging starts, the battery voltage starts to increase slowly, the battery surface temperature starts to increase slowly, and the battery voltage reaches the cut-off voltage at the late stage of constant current charging, at which time the battery surface temperature reaches the first peak, and the battery voltage is basically constant when charging at a constant voltage, and the surface temperature starts to decrease slowly. When close to the left side of the grating fiber of the sensor, the temperature deviation is greater, the highest deviation substance is about 0.96℃, and the overall maximum temperature difference is about 8.1℃. When the temperature deviation of the battery reaches 7.2 ℃, the strain on the surface of the battery begins to appear obvious aging state. At this time, the aging rate of the battery is the most obvious. When the discharge rate is increased to 2 ℃, the discharge rate of the surface increases by 50% through the fiber optic sensing and temperature measurement of the characteristics of the analysis. It can be seen that in the state of the constant charging rate and the discharge rate, the selection of different 5 positions for sensing monitoring, are able to show that the system has the high multiplexing capability, reliability and fast response.