Mixed visual angle integrated learning method for improving prediction precision of residual service life of lithium ion battery
Through the hybrid perspective integrated learning method, combined with OS-ELM, CEEMDAN and ARIMA, the health indicators and capacity degradation data perspective of lithium-ion batteries are integrated to construct meta-model data sets, which solves the problem of inaccurate prediction of single perspective in the existing technology and realizes the realization of the realization of inaccurate prediction of single perspectives. More accurate and reliable lithium-ion battery life prediction.
Patent Information
- Application Number
- CN202411040086.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-07-31
AI Technical Summary
Existing methods for predicting the residual service life of lithium-ion batteries usually rely on a single perspective and cannot fully capture the multifaceted characteristics of battery aging, resulting in inaccurate or incomplete predictions.
Adopting a hybrid perspective integrated learning method, health indicators and capacity degradation data are integrated through the online sequential extreme learning machine (OS-ELM) and fully adaptive noise ensemble empirical modal decomposition (CEEMDAN) algorithm combined with the autoregressive differential moving average model (ARIMA). Perspective, construct metamodel datasets for prediction.
It improves the accuracy and reliability of the residual service life prediction of lithium-ion batteries, enhances the robustness of the prediction model, and proves its advantages in improving prediction accuracy and reliability through experimental verification.
Smart Images

Figure CN118938057A_ABST
Abstract
Description
Technical Field
[0001] The present invention proposes a hybrid perspective ensemble learning method for improving the remaining useful life prediction accuracy of lithium-ion batteries, belonging to the technical field of RUL prediction of lithium-ion batteries. Background Art
[0002] As a source of energy for various electronic tools, lithium-ion batteries have become an indispensable part. However, due to the interference of internal response mechanisms and external environmental factors on battery charging and discharging, lithium-ion batteries will inevitably age, causing major safety accidents and serious consequences [1]. Therefore, it is extremely important to predict the remaining service life (RUL) of lithium-ion batteries.
[0003] The prediction methods for RUL of lithium-ion batteries mainly include model-based, data-driven and hybrid methods [2]. Liu et al. [3] proposed a dynamic latent variable reconstruction nonlinear Wiener process to predict RUL, but the electrochemical reaction inside the battery is very complex and difficult to model accurately. Amir et al. [4] proposed an equivalent circuit model that can adapt to and capture the dynamics of health status that changes over time to predict RUL. However, in practical applications, as the battery operating conditions change, the corresponding adjustment of model parameters may affect the accuracy of the model. Sangwan et al. [5] proposed an empirical model and Kalman filter to predict RUL. However, using this method to predict RUL will face the problem of difficult feature extraction.
[0004] Therefore, addressing complex problems in research requires considering multiple perspectives. Traditional approaches, while effective in some cases, may not fully grasp the complexity inherent in modern challenges. In response to this need for comprehensive solutions, the concept of a "hybrid perspective" has emerged as a promising paradigm. This approach advocates the integration of multiple viewpoints, methods, or data sources to achieve a more comprehensive understanding and solution to complex problems. Existing RUL prediction methods for lithium-ion batteries typically adopt a single perspective, either based on directly extracted HIs or on capacity degradation data of lithium-ion batteries. This can lead to a limited understanding of the complex dynamics of the battery degradation process and an inability to capture the multifaceted characteristics of battery aging, resulting in inaccurate or incomplete predictions.
[0005] In order to adopt the hybrid perspective method, this paper uses ensemble learning to combine two perspectives to predict the RUL of lithium-ion batteries. As a powerful technology, ensemble learning can integrate the advantages of each perspective, reduce the deviation and error of a single perspective, and thus improve the accuracy and reliability of the prediction. Applying ensemble learning to hybrid perspectives can organically integrate the information of hybrid perspectives, deepen the analysis of each perspective, and improve the effectiveness of the prediction model to meet various research challenges.
[0006] References:
[0007] [1] Zhao J, Zhu Y, Zhang B, Liu M, Wang J, Liu C and Hao X2023Sustainability 15 5014.
[0008] [2]Jiahui Z,Liting T et al.2023Power Generation Technology 44 1.
[0009] [3] Liu Q, Zhang Y, Si X and Fan Z 2023IEEE Transactions on Instrumentation and Measurement 72 1–9.
[0010] [4] Amir S, Gulzar M, Tarar MO, Naqvi IH, Zaffar NA and Pecht M G2022IEEE Access 10 18279–18288.
[0011] [5]Jafari S and Byun YC 2022Sensors 22 9522. Summary of the invention
[0012] The present invention proposes a hybrid perspective ensemble learning method (HyPELS) to improve the remaining useful life prediction accuracy of lithium-ion batteries to predict the RUL of lithium-ion batteries. First, the original data set obtained from the HIs perspective of the battery is shuffled and trained for capacity prediction through an online sequential extreme learning machine (OS-ELM). Secondly, the data from the battery capacity degradation data perspective is decomposed and reconstructed using the fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) algorithm, and then time series prediction and post-processing are performed through the autoregressive difference moving average model (ARIMA). Finally, in order to integrate the two perspectives, we propose a meta-model dataset construction method for training OS-ELM, which realizes RUL prediction from a hybrid perspective. Comprehensive experimental verification was carried out using two sets of NASA datasets, which fully demonstrated the effectiveness of HyPELS and proved its advantages in improving prediction accuracy and reliability.
[0013] The present invention provides a hybrid perspective ensemble learning method for improving the accuracy of remaining service life prediction of lithium-ion batteries, comprising the following steps:
[0014] Step 1: The original data set obtained from the perspective of battery HIs is shuffled and trained for capacity prediction through an online sequential extreme learning machine (OS-ELM).
[0015] Step 2: Decompose the data from the battery capacity degradation data perspective and reconstruct it using the fully adaptive ensemble empirical mode decomposition with noise (CEEMDAN) algorithm.
[0016] Step 3: Time series forecasting and post-processing through the Autoregressive In-Difference Moving Average (ARIMA) model.
[0017] Step 4: Integrate the two perspectives and propose a meta-model dataset construction method for training OS-ELM, realizing RUL prediction from a hybrid perspective.
[0018] The advantages and beneficial effects of the present invention are:
[0019] The present invention aims to improve the accuracy and robustness of lithium-ion battery RUL prediction by combining the perspectives of battery health indicators and capacity degradation data using an integrated learning method. Specifically, the main contributions of the present invention are as follows:
[0020] (1) Perspective integration: HyPELS effectively integrates the two perspectives of battery HIs and capacity degradation data through ensemble learning, making full use of the advantages of both to improve the accuracy and robustness of RUL prediction.
[0021] (2) Accuracy enhancement technology: HyPELS uses advanced technologies, such as shuffling the dataset during OS-ELM training to eliminate the impact of time, and uses complex methods to predict capacity degradation data to ensure the accuracy and reliability of the prediction results.
[0022] (3) Metamodel approach: HyPELS introduces a novel metamodel dataset construction method, which enables the OS-ELM model to be trained using merged datasets from multiple perspectives, facilitating seamless integration and enhancing the overall RUL prediction performance.
[0023] (4) Comprehensive experimental validation: The effectiveness of HyPELS is demonstrated through comprehensive experimental validation, which proves its advantages in improving prediction accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1a The capacity degradation curves of lithium-ion battery data B5 and B18 in the NASA constant voltage and constant current data set.
[0025] Figure 1b This is the capacity degradation curve of the NASA group's random walk charge and discharge data set RW9 and RW10.
[0026] Figure 2 A descriptive diagram for a mixed perspective (Perspective 1 is based on the HIs perspective of lithium-ion batteries, and Perspective 2 is based on the capacity degradation data perspective of lithium-ion batteries).
[0027] Figure 3a This is the algorithm flow chart of the fully adaptive noise ensemble empirical mode decomposition algorithm (CEEMDAN).
[0028] Figure 3b Figure 2 is the reconstruction process diagram of lithium-ion battery capacity degradation data after decomposition by CEEMDAN.
[0029] Figure 4 This is a graph of the four intrinsic mode components IMFs and a residual Res obtained after decomposition of NASA battery B5 using the CEEMDAN algorithm.
[0030] Figure 5 The present invention is a flowchart of the method.
[0031] Figure 6a The RUL prediction curve and actual curve diagram of the lithium-ion battery data B5 of the NASA constant voltage and constant current data set under the algorithm proposed in the present invention.
[0032] Figure 6b It is the RUL prediction curve and actual curve diagram of the lithium-ion battery data B18 of the NASA constant voltage and constant current data set under the algorithm proposed in the present invention.
[0033] Figure 6c The RUL prediction curve and actual curve diagram of the lithium-ion battery RW9 in the NASA group random walk charge and discharge data set under the algorithm proposed in the present invention.
[0034] Figure 6d It is the RUL prediction curve and actual curve diagram of the lithium-ion battery RW10 of the NASA group random walk charge and discharge data set under the algorithm proposed in the present invention.
[0035] Figure 7a To compare the RUL prediction curves and actual curves of the algorithm proposed in the present invention and other algorithms in the constant voltage and constant current data set of the NASA group lithium-ion battery data B5 from the HIs perspective in the experiment.
[0036] Figure 7b To compare the RUL prediction curves and actual curves of the algorithm proposed in the present invention and other algorithms in the HIs perspective NASA group constant voltage and constant current data set lithium-ion battery data B18 in the experiment.
[0037] Figure 7cTo compare the RUL prediction curves and actual curves of the algorithm proposed in the present invention and other algorithms for the random walk charge and discharge dataset RW9 of the NASA group from the HIs perspective in the experiment.
[0038] Figure 7d To compare the RUL prediction curves and actual curves of the algorithm proposed in the present invention and other algorithms for the random walk charge and discharge dataset RW10 of the NASA group from the HIs perspective in the experiment.
[0039] Figure 8a To compare the battery capacity degradation data in the experiment from the perspective of NASA group's constant voltage and constant current data set lithium-ion battery data B5, the RUL prediction curves and actual curves of the algorithm proposed in the present invention and other algorithms are shown.
[0040] Figure 8b To compare the battery capacity degradation data in the experiment, the NASA group constant voltage and constant current data set lithium-ion battery data B18 is used to show the RUL prediction curves and actual curves of the algorithm proposed in the present invention and other algorithms.
[0041] Figure 8c To compare the battery capacity degradation data in the experiment, the random walk charge and discharge dataset RW9 of the NASA group is used to show the RUL prediction curves and actual curves of the algorithm proposed in the present invention and other algorithms.
[0042] Figure 8d To compare the battery capacity degradation data in the experiment, the random walk charge and discharge dataset RW10 of the NASA group is used to show the RUL prediction curves and actual curves of the algorithm proposed in the present invention and other algorithms.
[0043] Figure 9a These are the RUL prediction curves and actual curves of the lithium-ion battery data B5 of the NASA group's constant voltage and constant current dataset in the ablation experiment in different integration methods (SAE, HYPELS) of a single perspective (Perspective1, Perspective2) and a mixed perspective. Figure 9b The RUL prediction curve and actual curve diagram of the lithium-ion battery data B18 of the NASA group constant voltage and constant current dataset in the ablation experiment in different integration methods (SAE, HYPELS) of a single perspective (Perspective1, Perspective2) and a mixed perspective. Fig.9c These are the RUL prediction curves and actual curves of the NASA group's random walk charge and discharge dataset RW9 in the ablation experiment in different integration methods (SAE, HYPELS) of a single perspective (Perspective1, Perspective2) and mixed perspective. Figure 9dThese are the RUL prediction curves and actual curves of the NASA group's random walk charge and discharge dataset RW10 in the ablation experiment in a single perspective (Perspective1, Perspective2) and different integration methods (SAE, HYPELS) of the mixed perspective. DETAILED DESCRIPTION
[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0045] The present invention provides a hybrid perspective ensemble learning method for improving the accuracy of remaining service life prediction of lithium-ion batteries, comprising the following steps:
[0046] Step 1: The original data set obtained from the perspective of battery HIs is shuffled and trained for capacity prediction through an online sequential extreme learning machine (OS-ELM).
[0047] Step 2: Decompose and reconstruct the data from the battery capacity degradation data perspective using the fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) algorithm;
[0048] Step 3: Time series forecasting and post-processing through the Autoregressive In-Difference Moving Average (ARIMA) model.
[0049] Step 4: Integrate the two perspectives and propose a meta-model dataset construction method for training OS-ELM, realizing RUL prediction from a hybrid perspective.
[0050] Among them, in step 1, specifically:
[0051] The two data sets used in the experiment come from NASA, collected from batteries with graphite anodes and lithium nickel cobalt manganese oxygen cathodes under two different experimental conditions. The first set of data sets was obtained by charging and discharging the battery under constant voltage and constant current conditions provided by NASA's Ames Prediction Center. The capacity degradation process of the battery is as follows Figure 1a As shown, they are B5 and B18 respectively. The 18650 lithium-ion battery was tested at 25°C, including three different operation modes: charge, discharge and impedance. The failure threshold was set to 70% of the rated capacity.
[0052] The second set of data sets comes from the random walk (RW) charge and discharge data set in NASA's open source database. Each set of 18650 lithium-ion batteries is operated with a continuous series of charge and discharge currents (ranging between -4.5A and 4.5A). This type of charge and discharge operation is called a random walk (RW) operation. The batteries used in this experiment are RW9 and RW10, and the failure threshold is set to 70% of the rated capacity. Their capacity degradation process is as follows Figure 1b shown.
[0053] like Figure 2 As shown, Figure 2 The front Perspective1 represents the HIs perspective, and the bottom Perspective2 represents the capacity degradation data perspective. Figure 2 The side is Figure 1a B5 and B18 battery capacity degradation curves. OS-ELM is an improved extreme learning machine (ELM) algorithm that overcomes the defect that ELM cannot perform online learning. The reason why OS-ELM performs better than ELM is that it adjusts the weights in each iteration, so that the network approaches the optimal solution faster. However, the current method of applying OS-ELM to lithium-ion battery RUL prediction usually directly divides the original data set into multiple groups according to time series, and then inputs them into the network for training in sequence. This practice causes the network to only learn from the data distribution of the first few cycles in the initial stage, and has no perception of the data distribution of other groups, which affects the network's learning quality for the overall data set. In other words, simply dividing the data set by time series to train OS-ELM will cause the network to be affected by the distribution of time series data, and it is impossible to purely consider HIs for learning, thereby introducing unnecessary interference terms, making it difficult for the network to converge to a better solution. In order to ensure that the training of OS-ELM is only related to HIs and not to the time series, it is necessary to shuffle the data set. Its formula is shown below.
[0054] D'=σ(D) (1)
[0055] Where D is the original dataset, D' is the shuffled dataset, and σ is the permutation function that rearranges the samples in dataset D.
[0056] OS-ELM consists of two phases: initialization phase and online sequential learning phase. In the initialization phase, OS-ELM is equivalent to ELM, where OS-ELM uses the first N0 training samples To train the network. i is the input vector of the ith sample, t i is the corresponding target output vector. Set the number of hidden layer nodes to k, the type of activation function to g(x), and minimize To obtain the optimal output weight matrix β0. Where w is the input weight, b is the hidden layer bias, β is the weight from the hidden layer to the output layer, and T0 is the target output matrix. Based on the above analysis, the processing process based on the HIs perspective is as follows.
[0057] 1) Randomly generate weight w i and bias b i ,in,
[0058] 2) Calculate the hidden layer output matrix H0.
[0059]
[0060] 3) Calculate the output weight matrix β (0) .
[0061]
[0062] in
[0063] 4) In the online sequential learning stage, N1 sample data are used for learning and the Get the most suitable output weight matrix β (1) .
[0064]
[0065] in,
[0066] 5) Substitute K1 into formula (4), and let The recursive formula for the online learning phase is:
[0067]
[0068] Through the above process, the shuffled dataset is passed through the trained OS-ELM to obtain the prediction results based on the HIs perspective.
[0069] Among them, in step 2, specifically:
[0070] In order to better understand and analyze the characteristics of signals, CEEMDAN is widely used in data preprocessing. The basic principle of CEEMDAN is to decompose complex signals into a set of intrinsic mode functions (IMFs) and a residual signal (Res). By introducing adaptive noise, CEEMDAN can more effectively suppress modal aliasing in each iteration, which not only improves the stability of signal decomposition, but also better handles nonlinear and non-stationary signals. The algorithm flow chart is shown in Figure 3aAs shown. Assume that x(t) is the original signal to be decomposed and δ(t) is Gaussian white noise. Repeatedly add Gaussian white noise with a mean of zero to the original signal for decomposition. Use empirical mode decomposition (EMD) to decompose to obtain the first intrinsic mode function (IMF) and residual (Res). Add specific noise to the residual of the first IMF for further decomposition to obtain the second IMF and residual. The EMD decomposition process continues until no further decomposition is possible. The formulas for the IMF and residual at the final stage are as follows.
[0071]
[0072] r d (t) = r d-1 (t)-IMF d (t) (7)
[0073] In formulas (6) and (7), IMF d (t) represents the dth modal component obtained by CEEMDAN decomposition, E d-1 (.) represents the d-1th modal component obtained by the EMD decomposition sequence, ε d-1 represents the weight coefficient of the residual noise added by CEEMDAN to the d-1th stage. d (t) represents the residual of the dth stage. Taking B5 as an example, the components obtained by decomposition are as follows Figure 4 shown.
[0074] The present invention decomposes the original capacity degradation data by CEEMDAN to obtain multiple IMFs and a Res. Then, the decomposed part is divided into high frequency and low frequency according to the zero crossing rate. For a signal sequence, if there are two adjacent sample values, one of which is positive and the other is negative, a zero crossing point will be generated. The formula of the zero crossing rate is as follows.
[0075]
[0076] Among them, n zero Indicates the number of signal zero crossings, s len Indicates the length of the signal, Rate zero-crossing Indicates the zero-crossing rate. If the zero-crossing rate is greater than or equal to 0.01, it is classified as the high-frequency part; if the zero-crossing rate is less than 0.01, it is classified as the low-frequency part.
[0077] The components obtained by CEEMDAN decomposition and zero-crossing rate need to be correctly reconstructed before being used for subsequent prediction. If not reconstructed, there will be redundant information and unnecessary memory consumption. Therefore, before prediction, we reconstruct the obtained high-frequency and low-frequency components according to different reconstruction rules to eliminate redundant information and improve prediction efficiency, such as Figure 3bshown.
[0078] (i) Low-frequency component. Low-frequency components refer to components with slower changes and longer periods that are related to the basic trend or main characteristics of the signal, and represent the long-term trend and periodic components in the original signal. In order to prevent the loss of important information and obtain low-frequency features, only a simple summation process is required for the low-frequency components.
[0079] (ii) The high-frequency part represents tiny fluctuations and transient events in the signal, which can help us understand the instantaneous state and local characteristics of the signal. When reconstructing, we must consider not only the global features but also the local features. Therefore, the present invention adopts two reconstruction rules to reconstruct the high-frequency part. One is to take the maximum value of each period of each high-frequency part as the value of the high-frequency feature of the period, which is called the maximum reconstruction rule. The other is to fully consider the joint influence of a point and its surrounding points on the data change trend, and use the window energy maximization rule to reconstruct the high-frequency part. The detailed steps of the window energy maximization reconstruction rule are as follows.
[0080] Assume that the decomposition produces m high-frequency parts Where e is the location of the data point in each high frequency part.
[0081] 1) Select a window of size 3*1 and calculate the energy value of each high-frequency part in the window area.
[0082]
[0083] In the formula, Indicates the energy value of each high-frequency part.
[0084] 2) The portion where the center point i of the window energy reaches the maximum value is taken as the high-frequency portion at position i.
[0085]
[0086] In the formula, max_value represents the maximum value of the window energy, and index represents the position index of the maximum value in the high-frequency part.
[0087] Among them, in step 3, specifically:
[0088] In the preprocessing of capacity degradation data, we obtained low-frequency and high-frequency features by decomposition and reconstruction. Due to the influence of capacity regeneration, there are usually mutations in low-frequency features, while high-frequency features show many short-term fluctuations and obvious non-stationarity. ARIMA can capture short-term fluctuations and mutations in data, and can also convert non-stationary signals into stationary signals through differential operations, so as to effectively model and predict. Therefore, ARIMA is used to analyze high and low frequency features. A brief description of ARIMA is as follows.
[0089] 1) Perform g-order differences on the non-stationary series until it becomes a stationary series. The formula for g-order differences is as follows.
[0090] Δ g y t =Δ(Δ g-1 y t ) (12)
[0091] Among them, the time series to be decomposed is y t , Δ represents the differential operation.
[0092] 2) Use AR to handle the autoregressive part of the time series.
[0093]
[0094] In the formula, y t represents the current value, μ is a constant term, and γ g is the autocorrelation coefficient, ε t is the error. p is the order of AR, which is determined by the partial autocorrelation coefficient function.
[0095] 3) Use MA to process the moving average part of the time series.
[0096]
[0097] In the formula, μ is a constant term, θ g is the autocorrelation coefficient, ε t is the error. q is the order of the MA, which is determined by the autocorrelation coefficient function.
[0098] A baseline value can be obtained by ARIMA analysis of low-frequency features that describe the capacity regeneration phenomenon and overall change trend, and two high-frequency features that represent non-stationary changes can be further analyzed to obtain a calibration value. By adding the baseline value to the two calibration values, we obtain two more accurate final results from the perspective of capacity degradation data.
[0099] Among them, in step 4, specifically:
[0100] We recognize that the two different perspectives differ not only in information features and model selection, but also in data preprocessing and prediction methods. It is challenging to integrate these two perspectives into ensemble learning. Therefore, we propose a meta-model dataset construction method for mixed-perspective ensemble learning, as follows. Assume that DR represents the original dataset of length L and DS represents the shuffled dataset.
[0101] 1) Obtain DS by shuffling DR. Randomly extract r samples from DS as training data set DS trainand record the index of these samples in DR as Idx DR (z), where z = 1, ..., s × r.
[0102] 2) Apply the idea of k-fold cross validation to convert DS train Divide into r subsets of equal size. Each subset is used as a validation set in turn, and the remaining r-1 subsets are combined into a training set. In this way, we get a combination of training and validation sets, denoted by C o , where o=1,..,r.
[0103] 3) From the perspective of HIs, traverse C o Perform model training and prediction to obtain s×r×1 groups of prediction results.
[0104] 4) From the perspective of capacity attenuation data, when traversing C o When, for each sample data, we in DS train Find its index z in , and thus locate its index Idx in DR DR (z). If Based on DR, we extract Idx DR (z) ARIMA time series forecasting is performed for several points before; on the contrary, if Based on DR, we reversely extract Idx DR (z) After post-processing, we get s×r×2 groups of prediction results.
[0105] 5) Combining the prediction results obtained from the two perspectives, we obtain a meta-model training dataset DM of size s×r×3 train .
[0106] Through the above steps, we integrate the perspectives of HIs and capacity degradation data into ensemble learning and finally obtain the DM for OS-ELM metamodel training. train , and finally the final prediction results are obtained through OS-ELM.
[0107] HyPELS uses ensemble learning to fuse the perspectives of HIs and capacity degradation data, taking full advantage of the advantages of both to alleviate the limitations of a single perspective and improve the accuracy and robustness of RUL prediction. In order to achieve accurate predictions from the HIs perspective, we shuffled the dataset to eliminate the influence of time factors in OS-ELM training. In order to achieve accurate predictions from the perspective of capacity degradation data, we performed data decomposition and reconstruction, ARIMA time series forecasting, and post-processing. In order to integrate the two perspectives, we proposed a meta-model dataset construction method to train OS-ELM, and finally achieved the final integration of RUL predictions from a mixed perspective. The overall algorithm flow chart of HyPELS is shown in the figure. Figure 5 shown.
[0108] In order to demonstrate the effectiveness and reliability of HyPELS, comparative experiments and ablation experiments were carried out using two sets of data sets. In the comparative experiment, the algorithms of the two perspectives were compared respectively. The actual curves and predicted curves of the battery capacity degradation data and the evaluation indicators of each algorithm are shown below, which proves the superiority of HyPELS. In the ablation experiment, ablation is gradually performed from the whole to the simple average ensemble (SAE) method and two single perspectives. The actual curves and predicted curves of the battery capacity degradation data and the evaluation indicators of each part are shown below, which proves the effectiveness of each perspective and integration method, thereby indicating the rationality of HyPELS.
[0109] The mean absolute error (MAE), root mean square error (RMSE) and coefficient of determination (R2) are introduced to measure the performance of HyPELS.
[0110]
[0111]
[0112] Among them, y k is the actual capacity of the lithium-ion battery, is the predicted capacity of the lithium-ion battery, is the average value of the actual capacity of lithium-ion batteries, and N is the number of predicted data.
[0113] In order to verify the feasibility of HyPELS, 70% of the initial capacity of the two groups of batteries was used as the failure threshold of the battery for the experiment. Table 1 shows the predicted value of the remaining service life of each group of lithium-ion batteries, and compares the predicted value and the true value of the remaining service life of each group of lithium-ion batteries. MAE, RMSE and R2 are used to evaluate the performance of different algorithms. Figure 6a , 6b , 6c and 6d show the fitting of the predicted curve of battery capacity degradation data with the actual curve.
[0114] Table 1 Prediction results of two groups of batteries by HyPELS
[0115]
[0116] As can be seen from Figure 6, the prediction curve of HyPELS for lithium-ion battery capacity degradation data is highly matched with the actual curve, and the predicted lithium-ion battery failure threshold (called RULpre) is close to the actual failure threshold (called RUL). Table 1 shows that RMSE and MAE are very small, while R2 is close to 1, indicating that HyPELS can predict RUL well, thus proving the rationality of HyPELS.
[0117] In order to demonstrate the advantages of HyPELS, this section compares the evaluation indicators of a series of algorithms based on capacity degradation data and HIs perspective. Figure 7a , Figure 7b , Figure 7c and Figure 7d The degree of fit between the predicted curve of each algorithm and the actual curve from the perspective of HIs is shown. Figure 8a , Figure 8b , Figure 8c and Figure 8d The fit between the predicted curve and the actual curve of each algorithm from the perspective of capacity degradation data is shown. The detailed comparison of evaluation indicators of each algorithm is shown in Table 2. Among them, Perspective1 represents the HIs perspective, and Perspective2 represents the capacity degradation data perspective shown in Table 3.
[0118] Table 2 Comparison of prediction results of different algorithms at each perspective and HyPELS
[0119]
[0120]
[0121] From the perspective of HIs, Figure 7a , Figure 7b , Figure 7c , Figure 7d As shown in Table 2, the MAE and RMSE of OS-ELM are the smallest among the methods under the same perspective, indicating that the error is the smallest. At the same time, the capacity prediction curve of each group of batteries has the best and most stable fit with the true capacity curve, which proves the rationality of the OS-ELM used. Figure 8a , Figure 8b , Figure 8c , Figure 8dAs shown in Table 2, from the perspective of capacity decay data, the CEEMDAN-ARIMA used in this experiment has the smallest predicted MAE and RMSE for each group of batteries, indicating that its error is much smaller than other algorithms from the same perspective, and the prediction accuracy is higher. At the same time, the RUL prediction curve of each group of batteries has the best fit with the actual capacity decay data curve from this perspective, indicating that this method has great advantages in predicting the capacity of lithium-ion batteries. Therefore, CEEMDAN-ARIMA is selected for capacity prediction from the perspective of capacity degradation data. Experimental data show that for the above two perspectives, the RUL capacity degradation data prediction curve obtained using HyPELS has a higher degree of fit with the actual curve, and the prediction error is significantly smaller than other methods, which proves the advantage of HyPELS.
[0122] To further demonstrate the effectiveness of HyPELS, we gradually perform ablations from the overall to the simple average ensemble (SAE) method and two single perspectives. In our experiments, we compare the single perspectives Perspective1 and Perspective2 and the two proposed ensemble methods SAE and HyPELS, using the evaluation criteria MAE, RMSE, and R2. Figure 9a , Figure 9b , Fig.9c , Figure 9d The capacity degradation data prediction curve and evaluation index comparison table of each method are given in Table 3.
[0123] Table 3 Comparison of prediction results of algorithms, SAE and HyPELS based on different quantitative perspectives
[0124]
[0125]
[0126] We use SAE to average the prediction results of Perspective1 and Perspective2. We use OS-ELM to integrate the prediction results of Perspective1 and Perspective2 into the prediction results of HyPELS. By comparing the evaluation indicators, it can be seen that the fit between the prediction curve of SAE capacity degradation data and the actual curve is better than the prediction result based on a single perspective. The prediction error is significantly smaller than the error of other methods based on a single perspective, which proves that HyPELS is indeed better than other prediction methods. HyPELS integrates two perspectives through OS-ELM. The fit between its capacity degradation data prediction curve and the actual curve is better than SAE, and the error is also smaller, which fully proves the effectiveness and superiority of HyPELS.
[0127] The HyPELS proposed in this invention considers both HIs and capacity degradation data, solving the limitation of traditional lithium-ion battery RUL prediction methods that rely on only a single perspective. This method can extract more comprehensive features and help obtain more accurate and reliable prediction results.
[0128] For the HIs perspective, considering the sensitivity of OS-ELM to time data, the dataset is shuffled to eliminate the impact of timing on model training and prevent unnecessary interference, thereby promoting the network to develop in the direction of a better solution. Subsequently, the prediction results of Perspective 1 are obtained using OS-ELM using the shuffled dataset.
[0129] For the capacity degradation data perspective, after decomposing it through CEEMDAN, the data is divided into high-frequency and low-frequency parts according to the zero-crossing rate. Then the obtained parts are reconstructed according to the reconstruction rules to obtain the reconstructed high-frequency and low-frequency features. Finally, ARIMA analysis and post-processing are used to obtain the prediction results of Perspective2.
[0130] In order to integrate the prediction results of two perspectives of HIs and capacity degradation data, the stacking method in ensemble learning is adopted. OS-ELM is trained as a meta-model. In order to facilitate the training of the meta-model, a set of metadata construction rules are designed to integrate the two perspectives of lithium-ion batteries. Two sets of NASA datasets are used to evaluate the performance of HyPELS. Comparative experiments and ablation experiments show that HyPELS exhibits lower error and higher accuracy compared with other methods, thus highlighting its significant advantages in the field of RUL prediction of lithium-ion batteries.
Claims
1. A hybrid perspective ensemble learning method for improving the accuracy of remaining useful life prediction of lithium-ion batteries, characterized by: The steps include: Step 1: The original data set obtained from the perspective of battery HIs is shuffled and capacity prediction training is performed through the online sequential extreme learning machine OS-ELM; Step 2: Decompose and reconstruct the data from the perspective of battery capacity degradation data using the fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) algorithm; Step 3: Time series forecasting and post-processing through the autoregressive difference moving average model ARIMA; Step 4: Integrate the two perspectives and propose a meta-model dataset construction method for training OS-ELM, realizing RUL prediction from a hybrid perspective.
2. A hybrid perspective ensemble learning method for improving the accuracy of remaining useful life prediction of lithium-ion batteries according to claim 1, characterized in that: In step 1, the two data sets used in the experiment come from NASA, collected from batteries with graphite anodes and lithium nickel cobalt manganese oxygen cathodes; The first set of data sets was obtained by charging and discharging the battery under constant voltage and constant current conditions provided by NASA's Ames Forecast Center; the second set of data sets came from the random walk charging and discharging data set in NASA's open source database.
3. The hybrid perspective ensemble learning method for improving the accuracy of remaining useful life prediction of lithium-ion batteries according to claim 1, characterized in that: In order to ensure that the training of OS-ELM is only related to HIs and not to time series, it is necessary to shuffle the dataset; the formula is as follows; D'=σ(D) (1) Where D represents the original data set, D' represents the shuffled data set, and σ represents the permutation function that rearranges the samples in the data set D.
4. A hybrid perspective ensemble learning method for improving the accuracy of remaining useful life prediction of lithium-ion batteries according to claim 1 or 3, characterized in that: OS-ELM consists of two phases: initialization phase and online sequential learning phase; In the initialization phase, OS-ELM uses the first N0 training samples To train the network; x i is the input vector of the ith sample, t i is the corresponding target output vector; the number of hidden layer nodes is set to k, the type of activation function is g(x), and by minimizing To obtain the optimal output weight matrix β0; where w is the input weight, b is the hidden layer bias, β is the weight from the hidden layer to the output layer, and T0 is the target output matrix.
5. A hybrid perspective ensemble learning method for improving the accuracy of remaining useful life prediction of lithium-ion batteries according to claim 4, characterized in that: The processing process based on the HIs perspective is as follows; 1.1 Randomly generate weight w i and bias b i ,in, 1.2 Calculate the hidden layer output matrix H0; 1.3 Calculate the output weight matrix β (0) ; in 1.4 In the online sequential learning stage, N1 sample data are used for learning and the Get the most suitable output weight matrix β (1) ; in, 1.5 Substitute K1 into formula (4), let The recursive formula for the online learning phase is; Through the above process, the shuffled dataset is passed through the trained OS-ELM to obtain the prediction results based on the HIs perspective.
6. The hybrid perspective ensemble learning method for improving the accuracy of remaining useful life prediction of lithium-ion batteries according to claim 1, characterized in that: In step 2, specifically: let x(t) be the original signal to be decomposed, δ(t) be Gaussian white noise; repeatedly add Gaussian white noise with a mean of zero to the original signal for decomposition; use empirical mode decomposition EMD to decompose to obtain the first intrinsic mode function IMF and residual Res; add specific noise to the residual of the first IMF for further decomposition to obtain the second IMF and residual; the EMD decomposition process continues until no further decomposition is possible; the formulas for the IMF and residual at the final stage are as follows; r d (t)=r d-1 (t)-IMF d (t) (7) In formulas (6) and (7), IMF d (t) represents the dth modal component obtained by CEEMDAN decomposition, E d-1 (.) represents the d-1th modal component obtained by the EMD decomposition sequence, εd -1 represents the weight coefficient of the residual noise added by CEEMDAN to the d-1 stage; r d (t) represents the residual of the dth stage.
7. A hybrid perspective ensemble learning method for improving the accuracy of remaining useful life prediction of lithium-ion batteries according to claim 6, characterized in that: The original capacity degradation data is decomposed by CEEMDAN to obtain multiple IMFs and a Res; then, the decomposed part is divided into high frequency and low frequency according to the zero crossing rate; if there are two adjacent sample values, one of which is positive and the other is negative, a zero crossing point will be generated; the formula of the zero crossing rate is as follows; Among them, n zero Indicates the number of signal zero crossings, s len Indicates the length of the signal, Rate zero-crossing Represents the zero-crossing rate; if the zero-crossing rate is greater than or equal to 0.01, it is classified as the high-frequency part; if the zero-crossing rate is less than 0.01, it is classified as the low-frequency part.
8. A hybrid perspective ensemble learning method for improving the accuracy of remaining useful life prediction of lithium-ion batteries according to claim 6 or 7, characterized in that: The steps of the window energy maximization reconstruction rule are as follows; Assume that the decomposition produces m high-frequency parts Where e is the location of the data point in each high-frequency part; 2.1 Select a window of size 3*1 and calculate the energy value of each high-frequency part in the window area; In the formula, Indicates the energy value of each high-frequency part; 2.2 The part with the center point i where the window energy has the maximum value is taken as the high-frequency part at position i; In the formula, max_value represents the maximum value of the window energy, and index represents the position index of the maximum value in the high-frequency part.
9. The hybrid perspective ensemble learning method for improving the accuracy of remaining useful life prediction of lithium-ion batteries according to claim 1, characterized in that: In step 3, ARIMA is used to analyze high and low frequency features. The steps are as follows; 3.1 Perform g-order differences on the non-stationary series until it becomes a stationary series; the formula for g-order differences is as follows; D g y t =D(D g-1 y t ) (12) Among them, the time series to be decomposed is y t , Δ represents the differential operation; 3.2 Use AR to process the autoregressive part of the time series; In the formula, y t represents the current value, μ is a constant term, and γ g is the autocorrelation coefficient, ε t is the error; p is the order of AR, which is determined by the partial autocorrelation coefficient function; 3.3 Use MA to process the moving average part of the time series; In the formula, μ is a constant term, θ g is the autocorrelation coefficient, ε t is the error; q is the order of MA, which is determined by the autocorrelation coefficient function.
10. The hybrid perspective ensemble learning method for improving the accuracy of remaining useful life prediction of lithium-ion batteries according to claim 1, characterized in that: In step 4, the specific steps are as follows: let DR represent the original data set of length L, and DS represent the shuffled data set; 4.1 Obtain DS by shuffling DR; randomly extract r samples from DS as the training data set DS train and record the index of these samples in DR as Idx DR (z), where z = 1, ..., s × r; 4.2 Apply k-fold cross validation to convert DS train Divide into r subsets of equal size; each subset is used as a validation set in turn, and the remaining r-1 subsets are combined into a training set; the combination of training and validation sets is obtained, denoted as C o , where o = 1, .., r; 4.3 From the perspective of HIs, traverse C o Perform model training and prediction to obtain s×r×1 sets of prediction results; 4.4 From the perspective of capacity attenuation data, when traversing C o When, for each sample data, in DS train Find its index z in , and thus locate its index Idx in DR DR (z); if Based on DR, extract Idx DR (z) ARIMA time series forecasting is performed for several points before; on the contrary, if Based on DR, reverse extraction of Idx DR (z) Perform ARIMA time series forecasting at several points after that; After post-processing, we obtained s×r×2 groups of prediction results; 5) The prediction results obtained from the two perspectives are combined to obtain a meta-model training dataset DM of size s×r×3 train ; Through the above steps, the perspectives of HIs and capacity degradation data are integrated into ensemble learning, and finally the DM for OS-ELM metamodel training is obtained. train , and finally the final prediction results are obtained through OS-ELM.
Citation Information
Patent Citations
Novel mixed lithium ion battery health state prediction method
CN113125987A
Lithium ion battery remaining service life prediction method and system based on IGS-SVM
CN113281671A
Lithium ion battery RUL prediction method based on HI and ANN
CN113627076A
Lithium battery life prediction method based on transfer learning and difference probability distribution
CN113791351A
Lithium ion battery remaining service life prediction method based on SAE-CEEMDAN-LSTM
CN114740360A
