A method for predicting SOH time series of lithium-ion power batteries for vehicles
By employing phase space reconstruction and a NARX neural network to capture the chaotic dynamics of lithium-ion batteries, the method addresses the limitations of existing SOH prediction methods, achieving improved accuracy and speed in lithium-ion power battery health status forecasting.
Patent Information
- Application Number
- CN202111062588.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-08
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2041-12-08
AI Technical Summary
Existing data-driven prediction methods for lithium-ion power battery health status (SOH) fail to account for the inherent nonlinear dynamic chaotic characteristics of the lithium battery system, resulting in insufficient information and reduced prediction accuracy.
A method involving phase space reconstruction technology to restore chaotic dynamics, using the correlation dimension D2 method to identify chaos, and a NARX neural network to predict SOH time series by determining embedding dimension and time delay, reconstructing one-dimensional data into a multidimensional state space for training.
Improves prediction accuracy by nearly 6 percentage points and convergence speed by 30 seconds, enhancing the NARX model's nonlinear prediction ability and response speed.
Abstract
Description
Technical field:
[0001] The present invention relates to a method for predicting parameter data of a lithium-ion power battery, and in particular to a method for predicting the SOH time series of a lithium-ion power battery for a vehicle. Background technology:
[0002] Lithium batteries have many advantages such as being green and environmentally friendly, having high specific energy and long service life. They are currently widely used in many fields such as new energy vehicles, mobile communications, smart grids and medical devices. However, with the frequent use of lithium-ion batteries, aging phenomena such as increased internal resistance and reduced capacity will occur, which will cause a series of safety problems and sometimes lead to catastrophic accidents, affecting the life safety of users. Therefore, accurate prediction of health status (SOH), state of charge (SOC) and other status quantities has great practical significance for improving the safety of lithium-ion power batteries.
[0003] At present, the research on the state of health (SOH) of lithium-ion power batteries is mainly based on data-driven prediction methods. For example, Xu Yuanzhong et al. used the simulated annealing algorithm (SA) to optimize the weights of the BP neural network in view of the local optimality of the weights of the BP network, so that the weights of the BP neural network are globally optimal. The experimental simulation effectively verified that the algorithm can effectively improve the prediction accuracy of the state of health (SOH) of lithium-ion power batteries; Zhang Ren et al. proposed a prediction algorithm based on particle swarm optimization RBF neural network, constructed an equivalent circuit model and thermodynamic model of lithium-ion power batteries, extracted key parameters affecting the SOH of lithium-ion batteries, and identified the key parameters online in real time. The results showed that the prediction accuracy was improved by 20% compared with the BP neural network, and the optimization time was saved by more than 66.7%; Pan Haihong and other scholars constructed a health indicator (HI) to characterize battery degradation, and then used the extreme learning machine algorithm to train and predict the health indicator, realizing the online prediction of the state of health (SOH), and the prediction error did not exceed 2%.
[0004] The above health state (SOH) prediction algorithms are mainly aimed at improving the shortcomings of traditional algorithms, and the data samples used are mostly time series or single variable data samples in one-dimensional state space. They have not studied the inherent nonlinear dynamic chaotic characteristics of the lithium battery system itself. Therefore, the information contained in the data samples is relatively simple and not comprehensive enough. Summary of the invention:
[0005] In view of the shortcomings of the prior art, the present invention provides a method for predicting the SOH time series of a lithium-ion power battery for a vehicle: by determining the chaotic dynamics of the lithium battery system, using the phase space reconstruction technology to reconstruct the time series data of the one-dimensional state space collected in the experiment, restoring the chaotic dynamic characteristic properties of the lithium battery system itself, and then using the NARX neural network to train and predict the reconstructed time series data to obtain the final SOH output value, including the following steps:
[0006] (1) Use the correlation dimension D2 method to identify the chaos in the SOH time series of lithium-ion power battery system;
[0007] (2) Establish a NARX prediction model;
[0008] (3) Collecting data sets from monitoring the degradation status of lithium-ion power batteries;
[0009] (4) Using grey correlation to select the input variables of the NARX prediction model from the collected data;
[0010] (5) The CC method is used to determine the embedding dimension and time delay. After calculation, τ = 3 and m = 8 are finally determined. The above results are used to reconstruct the phase space of the one-dimensional experimental data collected in the experiment, and the training sample set {(x i ,y i )},x i ∈R N×M ,y i ∈R N×1 , the test sample set {(x t ,y t )},x t ∈R N×M ,y t ∈R N×1 ;
[0011] (6) The reconstructed training sample set {(x i ,y i )},x i ∈R N×M ,y i ∈R N×1 Train the NARX model and set the test sample set {(x t ,y t )},x t ∈R N×M ,y t ∈R N×1 The NARX model is predicted to obtain the final SOH output value of the lithium-ion power battery.
[0012] Preferably, the correlation dimension D2 method in step (1) is as follows:
[0013] In the m-dimensional phase space, let the Euclidean distance between phase points be r. ij (m)
[0014] r ij (m) = ||X m (t i )-X m (t j )|| (1)
[0015] For any given scale r, the proportion of the number of point pairs smaller than r in all points is:
[0016]
[0017] Where n is the number of phase points, θ is the Heaviside function,
[0018] The correlation dimension D(m) is recorded as:
[0019]
[0020] Does not change with m, so the correlation dimension of the lithium-ion power battery system is obtained,
[0021]
[0022] When D2>2.3 is a fraction, it indicates that the lithium-ion power battery system has chaotic characteristics. Then the dimension of the attractor of the lithium-ion power battery system is D2, and the SOH time series of the lithium-ion power battery system can be predicted.
[0023] Preferably, the NARX network model in step (2) can be expressed as follows:
[0024] y(t)=f(y(t-1),y(t-2),…,y(tn y ),u(t-1),u(t-2),...,u(tn u )) (5)
[0025] Where: f is a nonlinear function; y(t) is the desired target vector; u(t) is the external input vector; y(t-1), y(t-2), ..., y(tn y ) is the expected target vector after the delay; u(t-1),u(t-2),...,u(tn u ) is the delayed external input vector, and the hidden layer activation function f of the NARX network model 1 Select tansig function; output layer activation function f 2 Select purelin function, which can be expressed as follows:
[0026]
[0027] f 2 (x) = x (7)
[0028] NARX neural network training uses the Levenberg-Marquards algorithm, assuming a maximum displacement as the region radius, and then looking for the minimum point of the cost function in the region. If the target cost function value increases, the region radius is adjusted to change the range and the solution is continued; if the target cost function value decreases, the iterative calculation is continued.
[0029] Preferably, the data set collected from monitoring the degradation state of the lithium-ion power battery in step (3) is a data set obtained by performing three experimental operations of charging, discharging and resistance measurement on the lithium-ion power battery at room temperature of 25° C., and measuring the battery terminal voltage, output current, temperature, charger voltage, charger current and other experimental data in real time, and the steps are as follows:
[0030] 1) Charge at 2A constant current until the battery voltage reaches 4.2V, then charge at constant voltage until I≤20mA;
[0031] 2) Discharge at a constant current of 2A until the voltages of batteries B5 and B6 drop to 2.7V and 2.5V respectively;
[0032] 3) Repeat the above steps (1) and (2) to cycle the charge and discharge process to age the lithium-ion battery. Stop the aging experiment when the actual capacity of the battery drops to 70% of the rated capacity (from 2Ah to 4Ah).
[0033] 4) The above steps (1) and (2) are recorded as one cycle process, and a total of 200 cycles are completed. The data acquisition frequency during the process is set to 1 Hz, and the battery terminal voltage, output current, temperature, charger voltage, charger current and data acquisition time are collected in real time and synchronously, and the maximum available capacity of the battery is recorded at the same time; the EIS method with a frequency scanning range of 0.1 Hz to 5 Hz is used to obtain the battery impedance.
[0034] Preferably, in step (5), the CC method determines the embedding dimension and time delay, and the steps are as follows:
[0035] 1) Assume τ s is the sampling interval of the lithium-ion power battery system time series, τ d =tτ s is the delay of the time series, τ w =(m-1)τ d is the delay time window, τ p is the average orbital period (τ w ≥τ p );
[0036] 2) Calculate the standard deviation σ of a given time series of a lithium-ion power battery system;
[0037] 3) S(t), ΔS(t) and S can be obtained by the following formula: cor (t) The value of the three variables, where r j =jσ / 2,(j=1、2、3、4)
[0038]
[0039]
[0040]
[0041] The three values from equation (8) to equation (10) comprehensively reflect the degree of correlation of time series data;
[0042] 4) Find the first local maximum independent of the time series from the first minimum of ΔS(t), by the time delay τ d =tτ s The first local maximum time can be obtained, and at the same time, S cor The minimum value of (t) determines the first overall maximum of the time series independent of the time window τ w =tτ s ; Using the above four steps, we can obtain τ=3 and m=8 for the lithium-ion battery dynamic system.
[0043] Preferably, the phase space reconstruction described in step (5) is based on the assumption that the lithium-ion power battery state of health (SOH) time series {x(i), i = 1, 2, ..., N}, according to Takens theory, if the time delay τ and the embedding dimension m have a certain value, then the lithium-ion power battery state of health (SOH) time series phase space phase point can be expressed as:
[0044]
[0045] Where m is the embedding dimension; τ is the time delay,
[0046] Y(j) is a point in the phase space, and the total number of phase points M satisfies the condition:
[0047] M=N-(m-1)τ (12)
[0048] For an M-dimensional multivariate time series X1,X2,...,X N , where any time series can be represented as X i =(x 1,i ,x 2,i ,...,x M,i), i = 1, 2, ..., N, when M = 1, it is a univariate sequence. Since the univariate time series can be considered as a special case of the multivariate time series, the phase point of the delayed phase space reconstruction of the multivariate time series is:
[0049]
[0050] in When m=m1+m2+…+m M , and m>2D (D is the attractor dimension), then there exists a mapping That is V i+1 =Φ(V i ), when m(or m1,m2,…,m M ) takes a sufficiently large value, the equivalent form of the mapping can be expressed as:
[0051] x 1,i+1 =Φ1(V i ),x 2,i+1 =Φ2(V i ),…,x M,i+1 =Φ M (V i ) (14)
[0052] The present invention discloses a method for predicting the SOH time series of a lithium-ion power battery for a vehicle, which uses the above algorithm to achieve the following beneficial effects:
[0053] The present invention utilizes the chaotic characteristics of a lithium-ion power battery system, adopts the CC method to determine the embedding dimension and time delay, and uses a NARX neural network to establish a lithium-ion power battery SOH time series prediction model. The one-dimensional input time series of the prediction model is reconstructed into a multidimensional state space through a phase space reconstruction technology, and the NARX neural network is used to train and predict the multidimensional state space time series to obtain a final SOH time series prediction value. The NARX prediction model of the present invention improves the mean square error of the RBF neural network by nearly 6 percentage points, and the convergence speed is increased by nearly 30 seconds. It has higher prediction accuracy and better response speed, and has higher application value. Description of the drawings:
[0054] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:
[0055] Figure 1 , a schematic diagram of the prediction model structure of NARX of the present invention;
[0056] Figure 2 , a schematic diagram of the SOH prediction results of the B5 battery of the present invention;
[0057] Figure 3 , a schematic diagram of the SOH prediction results of the B6 battery of the present invention;
[0058] Figure 4 , a schematic diagram of the relative error of the SOH prediction result of the B5 battery of the present invention;
[0059] Figure 5 , Schematic diagram of the average relative error of the prediction results of different algorithms for each battery;
[0060] Figure 6 , Schematic diagram of the root mean square error of the prediction results of different algorithms for each battery;
[0061] Figure 7 , Schematic diagram of the mean absolute error of the prediction results of different algorithms for each battery; Specific implementation method:
[0062] The present invention is further described in detail below in conjunction with the embodiments. The present invention discloses a method for predicting the SOH time series of a lithium-ion power battery for a vehicle: by determining the chaotic dynamics of the lithium battery system, using the phase space reconstruction technology to reconstruct the time series data of the one-dimensional state space collected in the experiment, restoring the chaotic dynamics characteristic properties of the lithium battery system itself, and then using the NARX neural network to train and predict the reconstructed time series data to obtain the final SOH output value, the steps are as follows:
[0063] (1) Use the correlation dimension D2 method to identify the chaos in the SOH time series of lithium-ion power battery system;
[0064] (2) Establish a NARX prediction model;
[0065] (3) Collecting data sets from monitoring the degradation status of lithium-ion power batteries;
[0066] (4) Using grey correlation to select the input variables of the NARX prediction model from the collected data;
[0067] (5) The CC method is used to determine the embedding dimension and time delay. After calculation, τ = 3 and m = 8 are finally determined. The above results are used to reconstruct the phase space of the one-dimensional experimental data collected in the experiment, and the training sample set {(x i ,y i )},x i ∈R N×M ,y i ∈R N×1 , the test sample set {(x t ,y t )},x t ∈R N×M ,y t ∈R N×1;
[0068] (6) The reconstructed training sample set {(x i ,y i )},x i ∈R N×M ,y i ∈R N×1 Train the NARX model and set the test sample set {(x t ,y t )},x t ∈R N×M ,y t ∈R N×1 The NARX model is predicted to obtain the final SOH output value of the lithium-ion power battery.
[0069] The correlation dimension D2 method in step (1) is as follows:
[0070] In the m-dimensional phase space, let the Euclidean distance between phase points be r. ij (m)
[0071] r ij (m) = ||X m (t i )-X m (t j )|| (1)
[0072] For any given scale r, the proportion of the number of point pairs smaller than r in all points is:
[0073]
[0074] Where n is the number of phase points, θ is the Heaviside function,
[0075] The correlation dimension D(m) is recorded as:
[0076]
[0077] Does not change with m, so the correlation dimension of the lithium-ion power battery system is obtained,
[0078]
[0079] When D2>2.3 is a fraction, it indicates that the lithium-ion power battery system has chaotic characteristics. Then the dimension of the attractor of the lithium-ion power battery system is D2, and the SOH time series of the lithium-ion power battery system can be predicted.
[0080] Among them, the NARX network model in step (2) can be expressed as follows:
[0081] y(t)=f(y(t-1),y(t-2),…,y(tn y ),u(t-1),u(t-2),…,u(tn u )) (5)
[0082] Where: f is a nonlinear function; y(t) is the desired target vector; u(t) is the external input vector; y(t-1), y(t-2),…, y(tn y ) is the expected target vector after the delay; u(t-1),u(t-2),…,u(tn u ) is the delayed external input vector, and the hidden layer activation function f of the NARX network model 1 Select tansig function; output layer activation function f 2 Select purelin function, which can be expressed as follows:
[0083]
[0084] f 2 (x) = x (7)
[0085] NARX neural network training uses the Levenberg-Marquards algorithm, assuming a maximum displacement as the region radius, and then looking for the minimum point of the cost function in the region. If the target cost function value increases, the region radius is adjusted to change the range and the solution is continued; if the target cost function value decreases, the iterative calculation is continued.
[0086] Among them, the experimental data set used in step (3) is mainly obtained from the NASA PCoE Research Center at the Idaho National Laboratory through the official website. There are 168 sets of data for B5 and B6 batteries respectively. In the independent experiment, two 18650 lithium-ion batteries B5 and B6 with a rated capacity of 2Ah were selected as experimental research objects. Three experimental operations such as charging, discharging and resistance measurement were performed at room temperature of 25°C, and experimental data such as battery terminal voltage, output current, temperature, charger voltage, and charger current were measured in real time.
[0087] The charge and discharge test method is to charge at a constant current of 1C (2A) until the battery voltage reaches 4.2V, and then charge at a constant voltage until I≤20mA; discharge at a constant current of 2A until the voltage of B5 and B6 batteries drops to 2.7V and 2.5V respectively; repeat the above charge and discharge cycle process to age the battery, and stop the aging experiment when the actual capacity of the battery drops to 70% of the rated capacity, that is, from 2Ah to 4Ah; the above operations are recorded as one cycle process, and the whole test completes 200 cycles in total. During the experiment, the data acquisition frequency is set to 1HZ, and the battery terminal voltage, output current, temperature, charger voltage, charger current and data acquisition time are collected in real time and synchronously, and the maximum available capacity of B5 and B6 batteries is recorded at the same time, and the battery impedance is obtained by the EIS method with a frequency scanning range of 0.1HZ~5HZ.
[0088] Among them, in step (5), the CC method determines the embedding dimension and time delay, and its steps are as follows:
[0089] 1) Assume τ s is the sampling interval of the lithium-ion power battery system time series, τ d =tτ s is the delay of the time series, τ w =(m-1)τ d is the delay time window, τ p is the average orbital period (τ w ≥τ p );
[0090] 2) Calculate the standard deviation σ of a given time series of a lithium-ion power battery system;
[0091] 3) S(t), ΔS(t) and S can be obtained by the following formula: cor (t) The value of the three variables, where r j =jσ / 2,(j=1、2、3、4)
[0092]
[0093]
[0094]
[0095] The three values from equation (8) to equation (10) comprehensively reflect the degree of correlation of time series data;
[0096] 4) Find the first local maximum independent of the time series from the first minimum of ΔS(t), by the time delay τ d =tτ s The first local maximum time can be obtained, and at the same time, S corThe minimum value of (t) determines the first overall maximum of the time series independent of the time window τ w =tτ s ; Using the above four steps, we can obtain τ=3 and m=8 for the lithium-ion battery dynamic system.
[0097] The phase space reconstruction described in step (5) assumes that the lithium-ion power battery state of health (SOH) time series {x(i), i = 1, 2, ..., N}, according to Takens theory, if the time delay τ and the embedding dimension m have a certain value, then the lithium-ion power battery state of health (SOH) time series phase space phase point can be expressed as:
[0098]
[0099] Where m is the embedding dimension; τ is the time delay,
[0100] Y(j) is a point in the phase space, and the total number of phase points M satisfies the condition:
[0101] M=N-(m-1)τ (12)
[0102] For an M-dimensional multivariate time series X1,X2,…,X N , where any time series can be represented as X i =(x 1,i ,x 2,i ,…,x M,i ), i = 1, 2, ..., N, when M = 1, it is a univariate sequence. Since the univariate time series can be considered as a special case of the multivariate time series, the phase point of the delayed phase space reconstruction of the multivariate time series is:
[0103]
[0104] in When m=m1+m2+…+m M , and m>2D (D is the attractor dimension), then there exists a mapping That is V i+1 =Φ(V i ), when m(or m1,m2,…,m M ) takes a sufficiently large value, the equivalent form of the mapping can be expressed as:
[0105] x 1,i+1 =Φ1(V i ),x 2,i+1 =Φ2(V i ),…,x M,i+1 =Φ M (V i ) (14)
[0106] The present invention selects the first 80 groups of data of B5 battery and the data monitored by the degradation state of lithium-ion battery using the independent test platform, and the training sample set obtained after phase space reconstruction is used as the training set of NARX neural network. The data monitored by the degradation state of lithium-ion battery using the independent test platform and the test set obtained after phase space reconstruction are used together with the last 88 groups of data as the test set of NARX neural network. The failure threshold is 0.7274. Figure 2 This is the SOH prediction result of B5 battery. Figure 4 Relative error of the SOH prediction result of B5 battery.
[0107] Similarly, the first 60 data sets of B6 batteries and the data monitored by the degradation state of lithium-ion batteries on the autonomous test platform were selected, and the training sample set obtained after phase space reconstruction was used as the training set of the NARX neural network. The test set obtained by monitoring the degradation state of lithium-ion batteries on the autonomous test platform and the phase space reconstruction was used together with the last 108 data sets as the test set of the NARX neural network. The failure thresholds of the B6 batteries were set to 0.68, Figure 3 This is the SOH prediction result of B5 battery.
[0108] Figure 5 is the average relative error of the prediction results of different algorithms for each battery, Figure 6 is the root mean square error of the prediction results of different algorithms for each battery, Figure 7 is the average absolute error of the prediction results of each battery by different algorithms. Figures 5 to 7 It can be seen that the relative error of the SOH prediction value of the NARX model is improved by nearly 6 percentage points compared with the RBF model, and the root mean square error is improved by nearly 5 percentage points, which further illustrates that the NARX model proposed in the present invention has higher accuracy and stability, and stronger nonlinear prediction ability.
[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the best embodiment, a person skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.
Claims
1. A method for predicting the SOH time series of automotive lithium-ion power batteries, by determining the chaotic dynamics of the lithium battery system, using the phase space reconstruction technology to reconstruct the time series data of the one-dimensional state space collected in the experiment, restoring the chaotic dynamics characteristic properties of the lithium battery system itself, and then using the NARX neural network to train and predict the reconstructed time series data, the steps are as follows: (1) Use the associated dimension D 2 The method is used to identify the chaos in the SOH time series of lithium-ion power battery system; (2) Establish a NARX prediction model; (3) Collecting data sets from monitoring the degradation status of lithium-ion power batteries; (4) Using grey correlation to select the input variables of the NARX prediction model from the collected data; (5) The CC method is used to determine the embedding dimension and time delay. After calculation, τ = 3 and m = 8 are finally determined. The determined embedding dimension and time delay are applied to the phase space reconstruction of the one-dimensional experimental data collected in the experiment, and the training sample set {(x i ,y i )},x i ∈R N×M ,y i ∈R N×1 , the test sample set {(x t ,y t )},x t ∈R N×M ,y t ∈R N×1 ; (6) The reconstructed training sample set {(x i ,y i )},x i ∈R N×M ,y i ∈R N×1 Train the NARX model and set the test sample set {(x t ,y t )},x t ∈R N×M ,y t ∈R N×I The NARX model is used to predict the final SOH output value of the lithium-ion power battery; The associated dimension D in step (1) 2 The method is as follows: In the m-dimensional phase space, let the Euclidean distance between phase points be r. ij (m) r ij (m)=||X m (t i )-X m (t j )|| (1) For any given scale r, the proportion of the number of point pairs smaller than r in all points is: Where n is the number of phase points, θ is the Heaviside function, The correlation dimension D(m) is recorded as: Does not change with m, so the correlation dimension of the lithium-ion power battery system is obtained, When D 2 >2.3 is a fraction, indicating that the lithium-ion power battery system has chaotic characteristics, so the dimension of the attractor of the lithium-ion power battery system is D 2 , predict the SOH time series of lithium-ion power battery system; The NARX network model in step (2) is expressed as follows: y(t)=f(y(t-1),y(t-2),…,y(t-n y ),u(t-1),u(t-2),…,u(t-n u )) (5) Where: f is a nonlinear function; y(t) is the expected target vector; u(t) is the external input vector; y(t-1), y(t-2), ..., y(tn y ) is the expected target vector after the delay; u(t-1), u(t-2), ..., u(tn u ) is the delayed external input vector, and the hidden layer activation function f of the NARX network model 1 Select tansig function; output layer activation function f 2 Select purelin function, which can be expressed as follows: f 2 (x)=x (7) NARX neural network training uses the Levenberg-Marquards algorithm, assuming a maximum displacement as the region radius, and then looking for the minimum point of the cost function in the region. If the target cost function value increases, the region radius is adjusted to change the range and continue to solve; if the target cost function value decreases, the iterative calculation continues; The data set collected from monitoring the degradation state of the lithium-ion power battery in step (3) is a data set obtained by performing three experimental operations of charging, discharging and resistance measurement on the lithium-ion power battery at room temperature of 25° C., and measuring the battery terminal voltage, output current, temperature, charger voltage and charger current experimental data in real time, and the steps are as follows: 1) Charge at 2A constant current until the battery voltage reaches 4.2V, then charge at constant voltage until I≤20mA; 2) Discharge at a constant current of 2A until the voltages of batteries B5 and B6 drop to 2.7V and 2.5V respectively; 3) Repeat the above steps 1) and 2) to cycle the charge and discharge process to age the lithium-ion battery. Stop the aging experiment when the actual capacity of the battery drops to 70% of the rated capacity, that is, from 2Ah to 4Ah; 4) The above steps 1) and 2) are recorded as one cycle process, and a total of 200 cycles are completed. The frequency of data acquisition during the process is set to 1HZ, and the battery terminal voltage, output current, temperature, charger voltage, charger current and data acquisition time are collected in real time and synchronously, and the maximum available capacity of the battery is recorded at the same time; the EIS method with a frequency scanning range of 0.1HZ to 5HZ is used to obtain the battery impedance; In step (5), the CC method determines the embedding dimension and time delay, and the steps are as follows: 1) Assume τ s is the sampling interval of the lithium-ion power battery system time series, τ d =tr s is the delay of the time series, τ w =(m-1)τ d is the delay time window, τ p is the average orbital period (τ w ≥τ p ); 2) Calculate the standard deviation σ of a given time series of a lithium-ion power battery system; 3) S(t), ΔS(t) and S are obtained by the following formula: cor (t) The value of the three variables, where r j =jσ / 2, (j=1, 2, 3, 4) The three values from equation (8) to equation (10) comprehensively reflect the degree of correlation of time series data; 4) Find the first local maximum independent of the time series from the first minimum of ΔS(t), by the time delay τ d =tr s Get the first local maximum time, and use S cor The minimum value of (t) determines the first overall maximum of the time series independent of the time window τ w =tτ s ; Using the above four steps, we can obtain τ=3, m=8 for the lithium-ion battery dynamic system; The phase space reconstruction described in step (5) assumes that the lithium-ion power battery state of health (SOH) time series {x(i), i = 1, 2, ..., N}, according to Takens theory, if the time delay τ and the embedding dimension m have a certain value, then the lithium-ion power battery state of health (SOH) time series phase space phase point is expressed as: Where m is the embedding dimension; τ is the time delay, Y(j) is a point in the phase space, and the total number of phase points M satisfies the condition: M=N-(m-1)τ (12) For an M-dimensional multivariate time series X 1 , X 2 , …, X N , where any time series is represented by X i =(x 1,i , x 2,i , …, x M,i ), i = 1, 2, ..., N, when M = 1, it is a univariate sequence. Since the univariate time series is a special case of the multivariate time series, the phase point of the delayed phase space reconstruction of the multivariate time series is: in i=1,2,…,M,when m=m 1 +m 2 +…+m M , and m>2D, D is the attractor dimension, then there exists a mapping That is V i+1 =Φ(V i ), when m or m 1 , m 2 ,…,m M When the value is large enough, the equivalent form of the mapping is expressed as: x 1,i+1 =Φ 1 (V i ),x 2,i+1 =Φ 2 (V i ),…,x M,i+1 =Φ M (V i ) (14)。
Citation Information
Patent Citations
Electric forklift lithium ion battery health state prediction method based on LSTM-FFNN
CN111308375A
Lithium battery SOH prediction method based on FWA optimization extreme learning machine
CN111950205A
Method and device for predicting SOH of battery
CN112924886A
KR20210036642A
