Lithium battery SOH and RUL collaborative prediction model based on multi-source feature selection and hybrid secondary mode decomposition

Abstract

The invention discloses a lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid secondary mode decomposition, and particularly relates to the field of battery performance prediction. Comprising the steps of obtaining a charge-discharge cycle curve of a to-be-predicted battery; extracting indirect health factors based on the charge-discharge cycle curve, namely a charge-related health factor, a discharge-related health factor and a hidden layer-related health factor; normalizing the capacity of the battery to be predicted and the indirect health factors, and screening an optimal health factor from the indirect health factors; performing modal decomposition on the optimal health factor to obtain a decomposition component; evaluating the decomposition component, and when an evaluation result reaches a preset condition, performing secondary modal decomposition on the decomposition component to obtain a modal component; and inputting the modal component into the prediction model for prediction, and carrying out reverse normalization on a prediction result to obtain the capacity and the remaining service life of the battery to be predicted. And the model prediction precision, generalization ability and robustness are improved.

Description

Technical Field

[0001] The present application relates to the field of battery health prediction, and in particular to a lithium battery SOH and RUL collaborative prediction model based on multi-source feature selection and hybrid quadratic modal decomposition. Background Art

[0002] As a typical energy device, lithium-ion batteries (LIBs) have the advantages of high specific energy, strong chargeability, flexible configuration, precise control and wide temperature range adaptability (-20 ℃~60 ℃). They are widely used in electric vehicles, multi / all-electric aircraft, ship electric propulsion systems, grid energy storage and other fields. Lithium-ion batteries are essentially a multi-scale, multi-physics, and multi-dimensional dynamic time-varying electrochemical system with complex internal reaction mechanisms, involving complex working conditions such as SEI film decomposition, lithium ion deposition, active material loss, overcharge, over-discharge, and mechanical abuse. Its performance degradation process is uncertain, nonlinear, irreversible and difficult to characterize, resulting in a significant reduction in equipment safety and reliability. Therefore, accurate prediction of lithium-ion battery SOH and RUL can effectively extend the battery cycle life, improve the robustness, reliability and safety of the battery management system (BMS), and ensure that the battery always operates at optimal efficiency. In addition, battery SOH and RUL prediction technology plays a key supporting role in the development of big data, blockchain and cloud platforms.

[0003] At present, health management based on lithium-ion batteries is mainly divided into two categories: direct health factor method and indirect health factor method. The direct health factor method directly estimates the state of indicators such as battery capacity, SOH, state of charge (SOC), state of energy (SOE) and state of power (SOP) by monitoring battery capacity, internal resistance and other characteristics offline. However, it has the disadvantages of difficulty in online monitoring of internal parameters, high calculation cost, poor model robustness, etc., and has no engineering practice value. In addition, the health factor of lithium-ion batteries is a feature vector that characterizes the battery SOH and RUL, and its extraction process largely determines the performance of downstream tasks of the data model. The existing indirect health factor method extracts too single feature types and the feature extraction process is complicated, which cannot fully reflect the battery capacity degradation trend, resulting in low model prediction accuracy. Summary of the invention

[0004] The main purpose of this application is to provide a lithium battery SOH and RUL collaborative prediction model based on multi-source feature selection and hybrid quadratic modal decomposition, aiming to solve the problem of low prediction accuracy of the existing indirect health factor method.

[0005] To achieve the above-mentioned objectives, the present application provides a lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid secondary modal decomposition, including: obtaining a charge and discharge cycle curve of a battery to be predicted; extracting a charge-related health factor, a discharge-related health factor and a hidden layer-related health factor based on the charge and discharge cycle curve; normalizing the capacity, charge-related health factor, discharge-related health factor and hidden layer-related health factor of the battery to be predicted, respectively, and filtering the normalized charge-related health factor, discharge-related health factor and hidden layer-related health factor, respectively; screening the filtered charge-related health factor, discharge-related health factor and hidden layer-related health factor to obtain an optimal health factor; performing modal decomposition on the optimal health factor to obtain a decomposition component; evaluating the decomposition component, and when it is determined that the evaluation result of the decomposition component meets a preset condition, performing secondary modal decomposition on the decomposition component to obtain a modal component; inputting the modal component into a prediction model for prediction to obtain a prediction result, and denormalizing the prediction result to obtain the capacity and remaining service life of the battery to be predicted.

[0006] Optionally, the charge and discharge cycle curve includes a battery charge and discharge voltage cycle curve, a battery charge and discharge current cycle curve, and a battery charge and discharge capacity cycle curve.

[0007] Optionally, extracting charge-related health factors, discharge-related health factors and hidden layer-related health factors based on the charge and discharge cycle curves includes: extracting incremental capacity curves, charge differential voltage curves and discharge differential voltage curves from the battery charge and discharge voltage cycle curves, the battery charge and discharge current cycle curves and the battery charge and discharge capacity cycle curves; filtering the hidden layer information in the incremental capacity curves, the charge differential voltage curves and the discharge differential voltage curves respectively; determining the charge-related health factors, discharge-related health factors and hidden layer-related health factors according to the filtered incremental capacity curves, the charge and discharge differential voltage curves and the capacity attenuation curves.

[0008] Optionally, the charging-related health factors include equal-voltage rise charging time, equal-interval charging voltage difference, constant-current charging time, constant-voltage charging time, equal-interval charging current difference and charging temperature peak time; the discharge-related health factors include equal-voltage drop discharge time, discharge voltage minimum point time, equal-interval discharge voltage difference, discharge temperature peak time, equal-interval discharge temperature difference, equal temperature difference time interval, constant-current discharge time, average discharge voltage and discharge current minimum point time; the hidden layer-related health factors include the peak intensity of the incremental capacity curve, the constrained dynamic peak area of ​​the characteristic peak of the incremental capacity curve, the unconstrained peak area of ​​the characteristic peak of the incremental capacity curve, the charging differential curve tolerance and the discharge differential curve peak intensity.

[0009] Optionally, screening among the charging-related health factors, the discharging-related health factors and the hidden layer-related health factors to obtain the optimal health factor includes: respectively determining the Pearson correlation coefficient and the Kendall correlation coefficient of the charging-related health factors, the discharging-related health factors and the hidden layer-related health factors; and screening out the related health factor with the largest Pearson correlation coefficient and the Kendall correlation coefficient as the optimal health factor.

[0010] Optionally, performing modal decomposition on the optimal health factor to obtain decomposition components includes: using a multivariate variational modal decomposition model to decompose the optimal indirect health factor to obtain decomposition components, wherein the decomposition components include trend components, intrinsic mode functions, and residual components.

[0011] Optionally, the decomposition components are evaluated, and when it is determined that the evaluation results of the decomposition components meet preset conditions, the decomposition components are subjected to secondary modal decomposition to obtain modal components, including: respectively determining the sample entropy, fuzzy entropy, approximate entropy and permutation entropy of each decomposition component; and when it is determined that the sample entropy, fuzzy entropy, approximate entropy and permutation entropy of all decomposition components are greater than a preset threshold, the residual components are subjected to secondary modal decomposition using an improved fully adaptive noise set empirical mode decomposition model to obtain modal components.

[0012] Optionally, after respectively determining the sample entropy, fuzzy entropy, approximate entropy and permutation entropy of each decomposition component, the method further includes: if the sample entropy, fuzzy entropy, approximate entropy and permutation entropy of any decomposition component are less than a preset threshold, taking the decomposition component as a modal component; inputting the modal component into the prediction model for prediction, and obtaining the prediction result includes: inputting the trend component, the intrinsic mode function and the residual component into the prediction model for prediction, obtaining the corresponding component prediction results, and summing all the component prediction results to obtain the prediction result.

[0013] Optionally, the prediction model is a deep belief network model.

[0014] To achieve the above-mentioned purpose, the present application also provides a lithium battery SOH and RUL collaborative prediction model based on multi-source feature selection and hybrid quadratic mode decomposition, which is characterized by comprising a cycle curve acquisition module, a health factor extraction module, a normalization module, a screening module, a decomposition module and a prediction module, wherein the cycle curve acquisition module is used to acquire the charge and discharge cycle curve of the battery to be predicted; the health factor extraction module is used to extract the charge-related health factor, the discharge-related health factor and the hidden layer-related health factor based on the charge and discharge cycle curve; the normalization module is used to respectively extract the capacity, charge-related health factor, discharge-related health factor and hidden layer-related health factor of the battery to be predicted. The normalized charging-related health factors, discharging-related health factors and hidden-layer-related health factors are normalized; the screening module is used to screen the normalized charging-related health factors, discharging-related health factors and hidden-layer-related health factors to obtain the optimal health factor; the decomposition module is used to perform modal decomposition on the optimal health factor to obtain decomposition components; the decomposition components are evaluated, and when it is determined that the evaluation results of the decomposition components meet the preset conditions, the decomposition components are subjected to secondary modal decomposition to obtain modal components; the prediction module is used to input the modal components into the prediction model for prediction to obtain prediction results, and the prediction results are denormalized to obtain the capacity and remaining service life of the battery to be predicted.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The lithium battery SOH and RUL collaborative prediction model based on multi-source feature selection and hybrid quadratic mode decomposition of the present invention normalizes the capacity, charge-related health factors, discharge-related health factors and hidden layer-related health factors of the battery to be predicted, ensures the maximum correlation between the battery capacity and the indirect health factors, and can realize the prediction of the battery capacity and the remaining service life at the same time; by organically combining the extraction process of indirect health factors of lithium-ion batteries with Pearson and Kendall correlation analysis, it reduces the difficulties of online monitoring of internal battery parameters and high calculation costs, and improves battery reliability and safety; in view of the complex internal degradation mechanism of the battery, the significant capacity decay and recovery phenomenon, a hybrid quadratic mode decomposition model is proposed, which uses the MVMD and ICEEMDAN models to comprehensively characterize the decay and non-stationarity of the capacity, and the best The information loss is avoided to the greatest extent; since the indirect health factor integrates multiple characteristic sources, it can effectively characterize the battery capacity attenuation through different voltage starting points, current starting points and sampling intervals, and the mixed quadratic mode decomposition subsequence of the optimal indirect health factor is used as the DBN model input to directly perform collaborative indirect prediction of the SOH and RUL of the lithium-ion battery, thereby improving the prediction accuracy, generalization ability and robustness of the model; the simultaneous output of RUL and SOH can provide more accurate battery status information, SOH can help monitor the health status of the battery in real time, while RUL provides a clearer time expectation for the battery life. The combination of the two can provide richer information for decision-making, help users optimize battery usage strategies, reduce costs, and improve the efficiency of battery use and maintenance. It is not only suitable for laboratory environments, but also for actual application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of a collaborative prediction method for SOH and RUL of lithium batteries based on multi-source feature selection and hybrid quadratic mode decomposition in this application; Figure 2 for Figure 1 A detailed flow chart of a method for obtaining a battery charge and discharge cycle curve; Figure 3 This is a schematic diagram of the structure of the deep belief network in the collaborative prediction method of SOH and RUL of lithium batteries based on multi-source feature selection and hybrid quadratic mode decomposition in this application; Figure 4 The RUL and SOH attenuation curves in the collaborative prediction method of SOH and RUL of lithium batteries based on multi-source feature selection and hybrid quadratic mode decomposition in this application; Figure 5 Extracting indirect health factor graphs from the collaborative prediction method of SOH and RUL of lithium batteries based on multi-source feature selection and hybrid quadratic mode decomposition in this application; Figure 6This is the heat map of the correlation of indirect health factors of B0005 battery in the collaborative prediction method of SOH and RUL of lithium batteries based on multi-source feature selection and hybrid quadratic mode decomposition in this application; Figure 7 This is the energy difference curve in the collaborative prediction method of SOH and RUL of lithium batteries based on multi-source feature selection and hybrid quadratic mode decomposition in this application; Figure 8 This is the decomposition result diagram of the MVMD algorithm in the collaborative prediction method of SOH and RUL of lithium batteries based on multi-source feature selection and hybrid quadratic mode decomposition in this application; Fig. 9 This is the evaluation result diagram of the lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid quadratic mode decomposition in this application; Fig.10 This is a graph of the decomposition results of ICEEMDAN in a collaborative prediction method for SOH and RUL of lithium batteries based on multi-source feature selection and hybrid quadratic mode decomposition in this application; Fig.11 The RUL and SOH collaborative prediction results of the B0005 battery at the 40th cycle in the lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid quadratic mode decomposition in this application; Fig.12 The RMSE results of B0005, B0006, B0007 and B0018 batteries at different cycle periods in the collaborative prediction method of SOH and RUL of lithium batteries based on multi-source feature selection and hybrid quadratic mode decomposition in this application; Fig.13 The MAE results of B0005, B0006, B0007 and B0018 batteries at different cycle periods in the collaborative prediction method of SOH and RUL of lithium batteries based on multi-source feature selection and hybrid quadratic modal decomposition in this application; Fig.14 The MAPE results of B0005, B0006, B0007 and B0018 batteries at different cycle periods in the collaborative prediction method of SOH and RUL of lithium batteries based on multi-source feature selection and hybrid quadratic mode decomposition in this application; Fig.15 The R of B0005 battery under different cycle periods in the collaborative prediction method of SOH and RUL of lithium battery based on multi-source feature selection and hybrid quadratic mode decomposition in this application is 2 result.

[0017] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of this application clearer, the technical solutions in this application will be clearly and completely described below in conjunction with the drawings in this application. Obviously, the described embodiments are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0019] The first embodiment of the present invention provides a lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid quadratic mode decomposition, such as Figure 1 As shown, the specific steps include: Step S1, obtaining a charge and discharge cycle curve of a battery to be predicted; Specifically, by simulating the capacity decay process of lithium-ion batteries and collecting and processing multi-source data of lithium-ion batteries in the process, the battery charge and discharge cycle curve is obtained. The battery charge and discharge cycle curve includes the battery charge and discharge voltage cycle curve, the battery charge and discharge current cycle curve, and the battery charge and discharge capacity cycle curve. The equipment used to obtain the battery charge and discharge cycle curve includes battery charge and discharge equipment, temperature recorder, battery comprehensive tester, electrochemical workstation, thermocouple, constant temperature and humidity chamber and host computer. Figure 2 As shown, the method for obtaining the battery charge and discharge cycle curve is as follows: Step S11, charging stage: the battery is placed in a 24 ℃ constant temperature and humidity chamber, and the battery charging and discharging equipment is used to charge the battery in a 1.5 A (0.75 C) constant current mode until the voltage reaches 4.2 V, and then the battery is charged in a 4.2 V constant voltage mode until the charging current drops to 20 mA (0.01 C), and the charging process ends; the battery surface temperature (T cha ) and sent to the host computer through the temperature recorder. The battery comprehensive tester collects the battery terminal voltage (V cha ), battery output current (I cha )、Charging time (t cha ), the voltage at the charger and the current at the charger, and send them to the host computer; Step S12, discharge phase: the ambient temperature is maintained at 24 °C, the battery is discharged in a constant current mode of 2 A (1 C), and the discharge process ends when the voltage of B5, B6, B7 and B18 lithium-ion batteries is lower than 24 °C. During the discharge process, the battery surface temperature (T cha ) and sent to the host computer through the temperature recorder. The battery comprehensive tester collects the battery terminal voltage (V dis ), battery output current (I dis )、Charging time (t dis ), capacity (Cap), load voltage and load current and send them to the host computer.

[0020] Step S13, impedance measurement: After the charge and discharge cycle is completed, the battery is allowed to stand for 30 minutes, and the electrochemical spectrum (EIS) in the range of 0.1 Hz to 5 kHz is used for frequency scanning. The battery comprehensive tester collects the battery AC impedance and sends it to the host computer, and monitors the change trend of its internal parameters with battery aging. The impedance collection data includes: estimated charge transfer resistance (Rct), estimated electrolyte resistance (Re), sensing branch current, battery branch current and battery impedance.

[0021] Step S14, repeat S11-S14 until the rated capacity of the battery drops to the failure threshold (1.4 A·h); the host computer establishes the battery charge and discharge voltage cycle curve, the battery charge and discharge current cycle curve, and the battery charge and discharge capacity cycle curve according to the collected data.

[0022] Step S2, extracting charging-related health factors, discharging-related health factors and hidden layer-related health factors based on the charge-discharge cycle curve; Step S2 specifically includes the following steps: Step S21, extracting an incremental capacity (IC) curve, a charge differential voltage (DV) curve, and a discharge differential voltage (DV) curve from a battery charge and discharge voltage cycle curve, a battery charge and discharge current cycle curve, and a battery charge and discharge capacity cycle curve; Step S22, high-order filtering is performed on the hidden layer information in the IC curve, the charging DV curve and the discharging DV curve respectively; wherein the filtering formula of the IC curve is:

[0023] In the formula, t, Q, I and V are the battery charge and discharge time, charge and discharge capacity, charge and discharge current and charge and discharge voltage respectively. k and k- 1 is the battery operation data at the current moment and the previous moment, detla_V and detla_Q are the unit voltage of IC curve and the unit capacity of DV curve respectively; The filtering formulas for the charge DV curve and the discharge DV curve are:

[0024] Step S23, determining indirect health factors according to the filtered IC curve, charging DV curve and discharging DV curve, specifically including charging-related health factors, discharging-related health factors and hidden layer-related health factors; specifically, the charging-related health factors include equal voltage rise charging time (HIs1), equal interval charging voltage difference (HIs2), constant current charging time (HIs3), constant voltage charging time (HIs4), equal interval charging current difference (HIs5) and charging temperature peak time (HIs6); the discharging-related health factors include equal voltage drop discharge time (HIs7), discharge voltage minimum point time (HIs8), equal interval discharge voltage difference (HIs9), and charging temperature peak time (HIs10). Is9), discharge temperature peak time (HIs10), equal interval discharge temperature difference (HIs11), equal temperature difference time interval (HIs12), constant current discharge time (HIs13), average discharge voltage (HIs14) and discharge current minimum point time (HIs15); hidden layer related health factors include IC curve peak intensity (HIs16), IC curve characteristic peak constrained dynamic peak area (HIs17), IC curve characteristic peak unconstrained peak area (HIs18), charging DV curve tolerance (HIs19) and discharge DV curve peak intensity (HIs20). The characteristic peak in this embodiment is determined according to actual conditions.

[0025] Step S3, specifically, normalize the battery capacity, charging-related health factor, discharge-related health factor, and hidden layer-related health factor. The normalization calculation formula is as follows:

[0026] Where: x is the original capacity; x norm is the normalized capacity.

[0027] Taking Peak2 as an example, the specific calculation formulas for various relevant health factors are introduced below, see Table 1.

[0028] Table 1 Indirect health factor extraction process

[0029] Where: Cha_V (i) is the battery charging voltage at time i; Cha_V (end) is the battery voltage when battery charging and discharging is terminated; Cha_I (i) Battery charging current at time i; Dis_V (i) is the battery discharge voltage at time i; Dis_I (i) is the battery discharge current at time i.

[0030] Step S4, screening the normalized charging-related health factors, discharging-related health factors and hidden layer-related health factors to obtain the optimal health factor; Step S41, respectively determine the Pearson correlation coefficient and Kendall correlation coefficient of the normalized charging-related health factor, discharging-related health factor, hidden layer-related health factor and normalized battery capacity; since the actual operation of the battery is affected by external factors such as environmental noise, the normalized indirect health factor of the battery must be filtered, and the Pearson correlation coefficient and Kendall correlation coefficient of the filtered indirect health factor and the normalized battery capacity are determined, and the correlation is compared by the Pearson correlation coefficient and the Kendall correlation coefficient, and the one with the greater correlation is used as the indirect health factor for subsequent steps; for example, if the correlation between the filtered indirect health factor and the normalized battery capacity is greater, the filtered indirect health factor is used as the indirect health factor, and if the correlation between the normalized indirect health factor and the normalized battery capacity is greater, the normalized indirect health factor is used as the indirect health factor. The results of whether each indirect health factor needs to be filtered are shown in Table 1. The filtering method can be a median filtering algorithm. The calculation formulas for the Pearson correlation coefficient and the Kendall correlation coefficient are as follows:

[0031]

[0032] Where X and Y are the indirect health factor and normalized capacity, respectively, X, Y∈[0, 1]; N is the battery cycle; c is the number of consistent pairs; d is the number of inconsistent pairs; 0.5n(n-1) is the binomial coefficient; ρ X,Y and τ a They are Person correlation coefficient and Kendall correlation coefficient, ρ X,Y , ρ X,Y ∈[-1, 1].

[0033] Step S42, screening out the relevant health factor with the largest Pearson correlation coefficient and Kendall correlation coefficient as the optimal health factor.

[0034] Step S5, modal decomposition is performed on the optimal health factor to obtain decomposition components, the decomposition components are evaluated, and when it is determined that the evaluation results of the decomposition components meet the preset conditions, secondary modal decomposition is performed on the decomposition components to obtain modal components. The specific steps are: Step S51, calculating the optimal decomposition mode number k according to the energy difference selection formula;

[0035] Where: E k The number of modal components is k is the sum of all component energies; θ is the energy difference; x l ( i ) is the modal decomposition under l The first component sequence i sampling points; n is the number of samples.

[0036] Based on the quadratic penalty factor, Lagrangian operator, augmented Lagrangian function, alternating direction multiplier and Wiener filtering, the optimal indirect health factor is decomposed into decomposition components using MVMD according to the optimal decomposition mode number k. The decomposition components include trend component, intrinsic mode function and residual component.

[0037] On the basis of the above steps, in order to further avoid the influence of capacity attenuation, capacity recovery and measurement noise on the battery RUL and SOH prediction results, the following steps are performed in this embodiment: Step S52, respectively determine the sample entropy, fuzzy entropy, approximate entropy and permutation entropy of the trend component, intrinsic mode function and residual component, and evaluate the complexity and self-similarity of the MVMD decomposition result. The calculation formulas are as follows:

[0038]

[0039]

[0040]

[0041] Step S53, if the evaluation results of the decomposed components meet the preset conditions, that is, the four entropy values ​​of the trend component, the intrinsic mode function and the residual component are all greater than the preset threshold value of 0.9, it is considered that the original signal contains multi-level and multi-scale features, and the improved completely adaptive noise ensemble empirical mode decomposition (ICEEMDAN) model is used to perform secondary mode decomposition on the residual component to obtain the modal component, which can avoid the shortcomings of strong randomness of high-frequency components and modal aliasing effects; If the sample entropy, fuzzy entropy, approximate entropy and permutation entropy of any decomposition component is less than the preset threshold, the decomposition component is taken as the modal component and the next step is directly performed; Step S6, training the DBN with the modal component as input and the normalized battery capacity as output; Step S7, input the modal component into the trained DBN for prediction to obtain a prediction result; perform denormalization on the prediction result to obtain the capacity and remaining service life of the battery. Further, the remaining number of cycles of the battery can be calculated based on the remaining service life.

[0042] Specifically, during the formal prediction, the input of the trained DBN is a modal component or a decomposition component, namely, a trend component, an eigenmode function, and a residual component; if the input is a modal component, the output is the prediction result; if the input is a decomposition component, the trend component, the eigenmode function, and the residual component are respectively input into the prediction model for prediction, and the corresponding component prediction results are obtained, and the prediction results of all components are summed up to obtain the prediction results. The prediction results are denormalized to obtain the capacity and remaining service life of the battery.

[0043] The prediction model is a deep belief network (DBN), which is a mixed probability generation directed model, which is composed of a graph structure restricted Boltzmann machine (RBM) and a Sigmoid belief network stacked together. The specific structure is as follows: Figure 3 shown.

[0044] Example 1 Based on the aging data of (PCoE) 18650 cylindrical lithium-ion batteries under different working conditions, the SOH and RUL collaborative indirect prediction of lithium-ion batteries is carried out. The specifications and external characteristic parameters of the tested batteries are shown in Table 1.

[0045] Table 1 Specifications and external characteristic parameters of the tested batteries

[0046] Obtain the battery charge and discharge voltage cycle curve, battery charge and discharge current cycle curve, and battery charge and discharge capacity cycle curve of four lithium-ion batteries B0005, B0006, B0007, and B0018, extract the IC curve, charge DV curve, and discharge DV curve from the battery charge and discharge voltage cycle curve, the battery charge and discharge current cycle curve, and the battery charge and discharge capacity cycle curve, and filter the above curves respectively. The original RUL and SOH attenuation curves of the battery are as follows: Figure 4 As shown, the SOH decay curve is calculated based on the utilization capacity. In order to better verify the feasibility and accuracy of the extraction and selection of multi-source indirect health factors, taking B0005 battery as an example, the calculation formula in Table 1 is used to extract indirect health factors according to the IC curve, charging DV curve and discharging DV curve. Figure 5 ,It can be seen from the figure that the extracted HIs8 is closest to the capacity curve.

[0047] Since the indirect health factor integrates multiple characteristic sources, it can effectively characterize the battery capacity decay through different voltage starting points, current starting points and sampling intervals. The correlation indicators between these indirect health factors and lithium-ion battery capacity determine the accuracy and effectiveness of the data model. Therefore, this embodiment calculates the Pearson correlation coefficient and Kendall correlation coefficient of each indirect health factor to obtain the B0005 battery indirect health factor correlation heat map, see Figure 6 As can be seen from the figure, the Person correlation coefficients of all health factors are greater than 0.98, and the Kendall correlation coefficients are greater than 0.86. In particular, the time when the discharge voltage reaches the lowest point (HIs8) has the strongest correlation with battery degradation, and its Person correlation coefficient and Kendall correlation coefficient are 0.99996 and 0.99357, respectively. Therefore, the time when the discharge voltage reaches the lowest point (HIs8) is selected as the optimal indirect health factor to directly perform hybrid synergistic prediction of battery RUL and SOH.

[0048] The optimal decomposition mode number K of the MVMD model is selected based on the energy difference. The energy difference curve of each mode component between VMD and MVMD algorithms is shown in Figure 7 As can be seen from the figure, the maximum reconstruction error and energy difference of the MVMD algorithm are much smaller than those of the VMD algorithm. When the number of decomposed modes is 4, the energy difference of the two algorithms is the smallest. Therefore, the optimal decomposition mode number K of the VMD and MVMD algorithms is 4. The normalized optimal indirect health factor is used as the input of the hybrid quadratic mode decomposition model, and the MVMD algorithm is used to decompose the optimal indirect health factor into trend component, intrinsic mode function and residual component. The results are shown in Figure 8 Then, the sample entropy, fuzzy entropy, approximate entropy and permutation entropy of each component are calculated, and the results are shown in Fig. 9 As shown in the figure, the four entropy values ​​of the trend component, intrinsic mode function and residual component are all greater than 0.9, proving that the original signal contains multi-level and multi-scale features. The ICEEMDAN model is used to perform secondary mode decomposition on the residual component. The results are shown in Fig.10 .

[0049] The decomposed residual components are input into DBN for prediction to obtain the prediction results, which are then denormalized to obtain the capacity and remaining service life of the battery.

[0050] Taking the data of the 40th cycle of B0005 as an example, the present invention uses the model of the present invention, VMD-PSO-DBN, PSO-DBN, DBN, CNN-LSTM, TCN, PSO-Bilstm, GA-SVM, GRU and RF models to collaboratively predict the RUL and SOH of B0005 batteries. The results are shown in Fig.11,As can be seen from the figure, GA-SVM, GRU and RF models have good ,prediction effects on the early RUL and SOH of lithium-ion batteries, but the ,prediction errors for the battery capacity degradation slope are large, which causes ,the synergistic prediction results to fluctuate and oscillate, and cannot accurately ,predict the battery end of life (EOL).

[0051] Since the battery has 7 capacity rebounds in the first 40% of its life cycle, the prediction results of the PSO-DBN, CNN-LSTM, DBN, TCN and PSO-Bilstm models are offset, reaching EOL 3, 4, 6, 7 and 10 cycles earlier, respectively. In order to better verify the prediction accuracy, generalization ability and robustness of the model, the comprehensive performance of the model is used, including RMSE, MAE, MAPE and R2. Figure 12-15 RMSE, MAE, MAPE and R of B0005, B0006, B0007 and B0018 batteries under different cycle periods 2 Results: As can be seen from the figure, the hybrid model of the present invention is superior to the nine comparison models under different evaluation indicators and has good prediction accuracy, generalization ability and robustness. Taking the prediction result of B5-40th as an example, compared with VMD-PSO-DBN, PSO-DBN, DBN, CNN-LSTM, TCN, PSO-Bilstm, GA-SVM, GRU and RF models, the RMSE of the model of the present invention is reduced by 43.67%, 48.33%, 52.24%, 74.09%, 78.03%, 79.53%, 79.9%, 80.48% and 82.31% respectively; the MAE is reduced by 34.91%, 39.13%, 47.68%, 70.23%, 75.74%, 76.59%, 78.37%, 79.59% and 80.17% respectively; the MAPE is reduced by 64.46%, 69.65%, 74.31%, 78.8%, 81.81%, 82.61%, 82.62% respectively. % and 82.64%. Similarly, the RMSE, MAE and MAPE of the model of the present invention are also effectively reduced at the 50th, 60th and 70th cycles. In addition, the R2 of the proposed model is 0.98786, while the R2 of the 9 comparative models are 0.97581, 0.9641, 0.95793, 0.85694, 0.80121, 0.81111, 0.76265, 0.74805 and 0.69331 respectively.

[0052] In summary, the model of the present invention is lower than the 9 comparative models in RMSE, MAE and MAPE indicators, and higher than the 9 comparative models in R2 indicators, and the maximum prediction error does not exceed 1 cycle, which proves that the proposed model has good nonlinear fitting ability and generalization ability, can effectively extend the battery cycle life, improve the robustness, reliability and safety of the battery management system (BMS), and ensure that the battery always operates at optimal efficiency.

[0053] The second embodiment of the present invention provides a lithium battery SOH and RUL collaborative prediction model based on multi-source feature selection and hybrid secondary modal decomposition, including a cycle curve acquisition module, a health factor extraction module, a screening module, a decomposition module and a prediction module, the cycle curve acquisition module is used to obtain the battery charge and discharge cycle curve; the health factor extraction module is used to extract the charge-related health factor, the discharge-related health factor and the hidden layer-related health factor based on the battery charge and discharge cycle curve; the normalization module is used to normalize the capacity, charge-related health factor, discharge-related health factor and hidden layer-related health factor of the battery to be predicted respectively; the screening module is used to screen the normalized charge-related health factor, discharge-related health factor and hidden layer-related health factor to obtain the optimal health factor; the decomposition module is used to evaluate the decomposition components, and when it is determined that the evaluation result of the decomposition component meets the preset condition, the decomposition component is subjected to secondary modal decomposition to obtain the modal component; the prediction module is used to input the decomposition component into the prediction model for prediction to obtain the prediction result, and the prediction result is denormalized to obtain the capacity and remaining service life of the battery.

[0054] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A collaborative prediction method for lithium battery SOH and RUL based on multi-source feature selection and hybrid quadratic mode decomposition, characterized in that: include: Obtaining the charge and discharge cycle curve of the battery to be predicted; Extracting a charge-related health factor, a discharge-related health factor and a hidden layer-related health factor based on the charge-discharge cycle curve; Normalizing the capacity of the battery to be predicted, the charging-related health factor, the discharging-related health factor, and the hidden layer-related health factor, respectively; Screening the normalized charging-related health factors, discharging-related health factors and hidden layer-related health factors to obtain the optimal health factor; Performing modal decomposition on the optimal health factor to obtain decomposition components; The decomposed components are evaluated, and when it is determined that the evaluation results of the decomposed components meet preset conditions, secondary modal decomposition is performed on the decomposed components to obtain modal components; The modal component is input into a prediction model for prediction to obtain a prediction result, and the prediction result is denormalized to obtain the capacity and remaining service life of the battery to be predicted.

2. The lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid quadratic modal decomposition according to claim 1 is characterized in that: The charge and discharge cycle curves include a battery charge and discharge voltage cycle curve, a battery charge and discharge current cycle curve, and a battery charge and discharge capacity cycle curve.

3. The lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid quadratic modal decomposition according to claim 2 is characterized in that: The extracting of the charging-related health factor, the discharging-related health factor and the hidden layer-related health factor based on the charging and discharging cycle curve includes: Extracting an incremental capacity curve, a charge differential voltage curve and a discharge differential voltage curve from the battery charge and discharge voltage cycle curve, the battery charge and discharge current cycle curve and the battery charge and discharge capacity cycle curve; filtering the hidden layer information in the incremental capacity curve, the charge differential voltage curve, and the discharge differential voltage curve respectively; The charge-related health factor, the discharge-related health factor and the hidden layer-related health factor are determined according to the filtered incremental capacity curve, the charge-discharge differential voltage curve and the capacity decay curve.

4. The lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid quadratic mode decomposition according to claim 1 is characterized in that: The charging-related health factors include equal voltage rise charging time, equal interval charging voltage difference, constant current charging time, constant voltage charging time, equal interval charging current difference and charging temperature peak time; The discharge-related health factors include equal voltage drop discharge time, discharge voltage minimum point time, equal interval discharge voltage difference, discharge temperature peak time, equal interval discharge temperature difference, equal temperature difference time interval, constant current discharge time, average discharge voltage and discharge current minimum point time; The hidden layer related health factors include the peak intensity of the incremental capacity curve, the constrained dynamic peak area of ​​the incremental capacity curve characteristic peak, the unconstrained peak area of ​​the incremental capacity curve characteristic peak, the charging differential curve tolerance and the discharging differential curve peak intensity.

5. The lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid quadratic mode decomposition according to claim 1 is characterized in that: The screening among the normalized charging-related health factors, discharging-related health factors and hidden layer-related health factors to obtain the optimal health factor includes: Determine the Pearson correlation coefficient and Kendall correlation coefficient of the normalized charging-related health factor, the discharging-related health factor, the hidden layer-related health factor and the normalized capacity of the battery to be predicted respectively; The relevant health factors with the largest Pearson correlation coefficient and Kendall correlation coefficient were selected as the optimal health factors.

6. The lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid quadratic mode decomposition according to claim 1 is characterized in that: The performing modal decomposition on the optimal health factor to obtain decomposition components includes: The optimal indirect health factor is decomposed by using a multivariate variational mode decomposition model to obtain decomposition components, which include trend components, intrinsic mode functions and residual components.

7. The lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid quadratic mode decomposition according to claim 6 is characterized in that: The step of evaluating the decomposed components and, when determining that the evaluation results of the decomposed components meet preset conditions, performing secondary modal decomposition on the decomposed components to obtain modal components includes: Determining the sample entropy, fuzzy entropy, approximate entropy and permutation entropy of each of the decomposed components respectively; When it is determined that the sample entropy, fuzzy entropy, approximate entropy and permutation entropy of all the decomposition components are greater than a preset threshold, the residual component is subjected to secondary modal decomposition using an improved fully adaptive noise set empirical mode decomposition model to obtain a modal component.

8. The lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid quadratic mode decomposition according to claim 7 is characterized in that: After respectively determining the sample entropy, fuzzy entropy, approximate entropy and permutation entropy of each of the decomposition components, the method further comprises: If the sample entropy, fuzzy entropy, approximate entropy and permutation entropy of any of the decomposition components are less than a preset threshold, the decomposition component is taken as a modal component; The step of inputting the modal component into a prediction model for prediction to obtain a prediction result comprises: The trend component, the intrinsic mode function and the residual component are input into a prediction model for prediction, and corresponding component prediction results are obtained. All component prediction results are summed to obtain a prediction result.

9. The lithium battery SOH and RUL collaborative prediction method based on multi-source feature selection and hybrid quadratic modal decomposition according to claim 1 is characterized in that: The prediction model is a deep belief network model.

10. A lithium battery SOH and RUL collaborative prediction model based on multi-source feature selection and hybrid quadratic mode decomposition, characterized in that: include: A cycle curve acquisition module is used to obtain the charge and discharge cycle curve of the battery to be predicted; A health factor extraction module, used to extract a charge-related health factor, a discharge-related health factor and a hidden layer-related health factor based on the charge-discharge cycle curve; A normalization module, used to normalize the capacity of the battery to be predicted, the charging-related health factor, the discharging-related health factor and the hidden layer-related health factor respectively; A screening module is used to screen the normalized charging-related health factors, discharging-related health factors and hidden layer-related health factors to obtain the optimal health factor; A decomposition module, used for performing modal decomposition on the optimal health factor to obtain decomposition components; The decomposed components are evaluated, and when it is determined that the evaluation results of the decomposed components meet preset conditions, secondary modal decomposition is performed on the decomposed components to obtain modal components; The prediction module is used to input the modal component into the prediction model for prediction, obtain the prediction result, and perform denormalization on the prediction result to obtain the capacity and remaining service life of the battery to be predicted.

Images