A Method for Predicting the Remaining Useful Life of Lithium Batteries Considering the Capacity Regeneration Phenomenon

Abstract

The present invention discloses a method for predicting the remaining useful life of a lithium battery considering the capacity regeneration phenomenon. By accelerating the life experiment of the lithium battery to be tested, the degradation amount of the lithium battery to be tested at different times is obtained; then, a capacity regeneration phenomenon detection algorithm is used to remove the capacity regeneration phenomenon in the historical degradation data of the lithium battery, and an LSTM network is trained with these processed degradation data; then, a capacity regeneration detection algorithm is used to detect the capacity regeneration phenomenon included in the lithium battery to be tested in real time, and the capacity regeneration data of the lithium battery to be tested is removed; finally, the trained LSTM neural network is used to perform real-time prediction of the RUL of the lithium battery to be tested, and the RUL prediction error caused by the capacity regeneration phenomenon is corrected, which has the characteristics of high prediction accuracy, good real-time performance, fast prediction speed, etc.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium battery reliability analysis, and more specifically, relates to a method for predicting the remaining useful life of a lithium battery considering the phenomenon of capacity regeneration. Background Art

[0002] In recent years, with the continuous development of electric vehicles, lithium batteries have been increasingly used in the power supply systems of electric vehicles. Therefore, for electric vehicles, the reliability of lithium batteries affects the stability of the entire operation of electric vehicles, and the research on the prediction technology method of the remaining useful life (RUL) of lithium batteries has become very necessary, which has the following important significances: (1) It is an important way to obtain the reliability information of electric vehicles and can further provide a basis for realizing online monitoring and health management of electric vehicles; (2) It helps to prompt manufacturers to improve the process of lithium battery modules, thereby introducing new materials and improving packaging technologies; (3) It is beneficial to better design accelerated aging tests to obtain more accurate lithium battery aging data; (4) It can realize condition-based maintenance, enabling end-users to obtain more life information of electric vehicles to reduce the investment in the maintenance of electric vehicles.

[0003] During the degradation process of lithium batteries, if the lithium batteries are in a non-working state for a long time, their battery capacity will recover to a certain extent, and this phenomenon is called the capacity regeneration phenomenon of lithium batteries. For the same type of lithium batteries, the RUL of lithium batteries without the capacity regeneration phenomenon is shorter than that of lithium batteries with the capacity regeneration phenomenon.

[0004] In the actual use process of lithium batteries, the switching between their working state and non-working state often depends on the user's usage habits, which results in the capacity regeneration phenomenon of lithium batteries often showing a certain randomness. Therefore, this randomly occurring capacity regeneration phenomenon is likely to cause the degradation model trained based on historical lithium battery data to be difficult to adapt to the degradation trajectory of the same type of target lithium batteries, resulting in inaccurate RUL prediction results. Most of the existing lithium battery RUL prediction technologies do not consider the impact of the capacity regeneration phenomenon of lithium batteries on the prediction results, but simply regard the recovery of lithium battery capacity as noise and error during the degradation process and use some noise reduction algorithms to process it. This processing method cannot accurately remove the error brought by the capacity regeneration phenomenon to the RUL prediction of lithium batteries, and still does not solve the difference in the degradation trajectories of the same type of lithium batteries caused by the capacity regeneration phenomenon. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a method for predicting the remaining useful life of a lithium battery considering the capacity regeneration phenomenon, which is used to eliminate the error brought by the capacity regeneration phenomenon to the RUL prediction of the lithium battery, improve the prediction accuracy, and thus ensure the reliability of the lithium battery operation.

[0006] To achieve the above-mentioned invention purpose, a method for predicting the remaining useful life of a lithium battery considering the capacity regeneration phenomenon according to the present invention is characterized by including the following steps:

[0007] (1) Obtain the capacity of the lithium battery at different times;

[0008] By accelerating the life experiments of N lithium batteries, sample the capacity of each lithium battery at different times. Among them, the capacity of the i-th lithium battery at different times is denoted as where i = 1, 2,..., N, represents the initial capacity of the i-th lithium battery, represents the capacity of the i-th lithium battery at the t-th moment, t = 0, 1, 2,... T, and T represents the number of sampling moments;

[0009] (2) Detect the capacity regeneration phenomenon in the historical data of the lithium battery;

[0010] (2.1) Obtain the first-order difference of the capacity data of the i-th lithium battery where

[0011] (2.2) Obtain the first-order difference data of the first k non-capacity-regeneration phenomena of the i-th lithium battery T > k > 2, and then calculate the mean μ i and the variance σ i :

[0012]

[0013]

[0014] (2.3) Let k = k + 1, and judge whether k is less than T. If so, enter step (2.4); if not, enter step (3);

[0015] (2.4) Judge whether is greater than min(0, μ i + 3σ i ). If not, return to step (2.2); if so, enter step (2.5);

[0016] (2.5) Set a counting variable q, whose initial value is 0; let q = q + 1, and then record the starting moment when the q-th capacity regeneration phenomenon occurs in the i-th lithium battery and the starting capacity corresponding to the starting moment and enter step (2.6);

[0017] (2.6) Let k = k + 1 again and judge whether it is less than If not, re-enter (2.6); if so, record the end moment when the qth capacity regeneration phenomenon occurs in the ith lithium battery and the end capacity corresponding to the end moment and enter step (2.7);

[0018] (2.7) Remove the capacity data i from the capacity data X of the ith lithium battery Then let and return to step (2.2);

[0019] (3) Train the long short-term memory neural network LSTM;

[0020] After the capacity data of N lithium batteries are processed by step (2), the historical capacity data of the lithium batteries without capacity regeneration phenomenon is obtained, denoted as where n is the number of each group of data; then use these N groups of data to train an LSTM neural network;

[0021] (4) Real-time prediction of the life of the lithium battery to be tested

[0022] (4.1) Obtain the capacity degradation data X = [x 0 , x 1 ... x t of the lithium battery to be tested at the current moment, and judge whether x t is less than the failure threshold w. If it is less, go to step (5) and the algorithm ends; otherwise, go to step (4.2);

[0023] (4.2) Calculate the first-order difference ΔX = [Δx 1 , Δx 2 ... Δx t of the data X, where Δx j = x j - x j-1 , j = 1, 2,..., t; then calculate the mean μ and variance σ;

[0024]

[0025]

[0026] (4.3) Judge Δx tIs it greater than min(0, μ + 3σ)? If not, enter (4.4); if so, enter (4.5);

[0027] (4.4), Input the data X = [x o , x 1 ... x t into the LSTM neural network to predict the remaining useful life RUL of the lithium battery to be tested at time t t , then let t = t + 1, and then enter step (4.1);

[0028] (4.5), Set a counting variable p with an initial value of 0; let p = p + 1, and then record the starting time t p，begin = t - 1 when the p-th capacity regeneration phenomenon of the lithium battery to be tested occurs, and the starting capacity corresponding to the starting time Calculate the remaining useful life corresponding to the current time t as where represents the predicted remaining useful life value at time t p，begin predicted by the LSTM neural network, and then enter step (4.6);

[0029] (4.6), Let t = t + 1 again, and judge whether x t is less than If not, calculate the predicted remaining useful life value at the current time as and re-enter (4.6); if so, record the ending time t p，end = t when the p-th capacity regeneration phenomenon of the lithium battery to be tested occurs, and calculate the predicted remaining useful life result at the current time t p，end as Then enter step (4.7);

[0030] (4.7), Add the correction value E = t p，end - t p，begin - 1 to all the predicted remaining useful life values obtained currently, and remove the data of the capacity regeneration phenomenon from the capacity degradation data X of the lithium battery to be tested Then let t = t p，begin + 1, and then return to step (4.1);

[0031] (5), When the capacity of the lithium battery to be tested degrades to the failure threshold w, the prediction terminates and the algorithm ends.

[0032] The invention object of the present invention is realized as follows:

[0033] A method for predicting the remaining useful life of a lithium battery considering the capacity regeneration phenomenon, which accelerates the life experiment of the lithium battery to be tested to obtain the degradation amount of the lithium battery to be tested at different times; then uses the capacity regeneration phenomenon detection algorithm to remove the capacity regeneration phenomenon in the historical degradation data of the lithium battery, and trains an LSTM network with these processed degradation data; then uses the capacity regeneration detection algorithm to detect the capacity regeneration phenomenon included in the lithium battery to be tested in real time and remove the capacity regeneration data of the lithium battery to be tested; finally, uses the trained LSTM neural network to perform real-time prediction of the RUL of the lithium battery to be tested and correct the RUL prediction error caused by the capacity regeneration phenomenon, which has the characteristics of high prediction accuracy, good real-time performance, and fast prediction speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a flowchart of a method for predicting the remaining useful life of a lithium battery considering the capacity regeneration phenomenon according to the present invention;

[0035] Figure 2 are 4 groups of lithium battery capacity degradation data obtained from the accelerated life experiment;

[0036] Figure 3 is the detection result of the capacity regeneration phenomenon detection algorithm;

[0037] Figure 4 are the degradation trajectories of the capacities of 4 groups of lithium batteries after removing the capacity regeneration phenomenon;

[0038] Figure 5 is a graph of the remaining useful life prediction result of a new method for predicting the remaining useful life of a lithium battery considering the capacity regeneration phenomenon proposed by the present invention:

[0039] Figure 6 are the results of predicting the remaining useful life of a lithium battery by three prediction models: (1) a new method for predicting the remaining useful life of a lithium battery considering the capacity regeneration phenomenon; (2) a remaining useful life prediction model based on the traditional Wiener process; (3) a remaining useful life prediction model based on the LSTM neural network. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] The following describes the specific embodiments of the present invention with reference to the drawings, so that those skilled in the art can better understand the present invention. It should be particularly noted that in the following description, when the detailed description of known functions and designs may obscure the main content of the present invention, these descriptions will be omitted here.

[0041] Embodiment

[0042] Figure 1 is a flowchart of a method for predicting the remaining useful life of a lithium battery considering the capacity regeneration phenomenon according to the present invention.

[0043] In this embodiment, asFigure 1 As shown in the figure, a method for predicting the remaining life of a lithium battery considering the capacity regeneration phenomenon according to the present invention includes the following steps:

[0044] S1. Obtain the capacity of the lithium battery at different times;

[0045] By accelerating the life experiments of N lithium batteries, sample the capacity of each lithium battery at different times. Among them, the capacity of the i-th lithium battery at different times is denoted as where i = 1, 2,..., N, represents the initial capacity of the i-th lithium battery, represents the capacity of the i-th lithium battery at the t-th time, t = 0, 1, 2,...T, and T represents the number of sampling times;

[0046] In this embodiment, lithium battery B0005 is selected as the experimental object, and the degradation data of a total of 3 groups of lithium batteries, namely B0006, B0007, and B0018, are selected as historical degradation data to simulate the prediction of the remaining life of the lithium battery under real-time working conditions. Figure 2 are the capacity degradation data of 4 groups of lithium batteries obtained from the experiment. The sampling time values of the 4 groups of lithium batteries are 166 cycles, 163 cycles, 166 cycles, and 132 cycles respectively; it can be clearly seen that Figure 2 the lithium battery degradation trajectories in contain a large number of capacity regeneration phenomena, and these capacity regeneration phenomena will greatly affect the accuracy of the RUL prediction model.

[0047] S2. Detect the capacity regeneration phenomenon in the historical data of the lithium battery;

[0048] S2.1. Obtain the first-order difference of the capacity data of the i-th lithium battery where

[0049] S2.2. Obtain the first-order difference data of the first k of the i-th lithium battery that do not contain the capacity regeneration phenomenon T > k > 2, in this embodiment, k = 5 is taken, and then the mean value μ is calculated i and the variance σ i :

[0050]

[0051]

[0052] S2.3. Let k = k + 1, and determine whether k is less than T. If so, go to step S2.4; if not, go to step S3;

[0053] S2.4. Determine whether is greater than min(0, μi +3σ i ), if not, return to step S2.2; if so, go to step S2.5;

[0054] S2.5. Set a counting variable q with an initial value of 0; let q = q + 1, and then record the starting time when the q-th capacity regeneration phenomenon occurs in the i-th lithium battery and the starting capacity corresponding to the starting time and enter step S2.6;

[0055] S2.6. Let k = k + 1 again and judge whether it is less than If not, execute S2.6 again; if so, record the ending time when the q-th capacity regeneration phenomenon occurs in the i-th lithium battery and the ending capacity corresponding to the ending time and enter step S2.7;

[0056] S2.7. Remove the capacity data i from the capacity data X of the i-th lithium battery and then let and return to step S2.2;

[0057] In this embodiment, the capacity regeneration detection algorithm described in steps S2.1 - S2.7 is used to adaptively detect the capacity regeneration phenomenon of the degradation data of 3 groups of lithium batteries, and the results are as Figure 3 shown. The lithium battery has multiple capacity regeneration phenomena during the entire experimental stage. Then, according to the detection results of the capacity regeneration detection algorithm, excluding the capacity regeneration phenomenon during the degradation process of the lithium battery, the lithium battery degradation trajectory without the capacity regeneration phenomenon is as Figure 4 shown.

[0058] S3. Train a long short-term memory neural network LSTM;

[0059] After the capacity data of N lithium batteries are processed by step S2, the historical capacity data of the lithium battery without the capacity regeneration phenomenon is obtained, denoted as where n is the number of each group of data; then use these N groups of data to train an LSTM neural network;

[0060] In this embodiment, the degradation data of lithium batteries B0006, B0007, and B0018 with the capacity regeneration phenomenon removed are used to train an LSTM neural network, and the specific training process of the LSTM neural network will not be elaborated here.

[0061] S4. Real-time prediction of the life of the lithium battery to be measured

[0062] S4.1. Obtain the capacity degradation data X = [x 0 , x 1 ... x t of the lithium battery B0005 to be measured at the current moment, and determine whether x t is less than the failure threshold w = 1.35 Ah. If it is less, go to step S5 and the algorithm ends; otherwise, go to step S4.2;

[0063] S4.2. Calculate the first-order difference ΔX = [Δx 1 , Δx 2 ... Δx t of the data X, where Δx j = x j - x j-1 , j = 1, 2,..., t; then calculate the mean μ and variance σ;

[0064]

[0065]

[0066] S4.3. Determine whether Δx t is greater than min(0, μ + 3σ). If not, enter S4.4; if so, enter S4.5;

[0067] S4.4. Input the data X = [x 0 , x 1 ... x t into the LSTM neural network to predict the remaining useful life RUL t of the lithium battery to be measured at time t, then let t = t + 1, and then enter step S4.1;

[0068] S4.5. Set the counting variable p with its initial value of 0; let p = p + 1, and then record the starting time t p，begin = t - 1 when the p-th capacity regeneration phenomenon of the lithium battery to be measured occurs, and the starting capacity corresponding to the starting time Calculate the remaining useful life corresponding to the current time t as where represents the remaining useful life prediction value at time t p，begin predicted by the LSTM neural network, and then enter step S4.6;

[0069] S4.6. Let t = t + 1 again, and determine whether x t is less than If not, calculate the remaining useful life prediction value at the current time as and re - execute S4.6; if so, record the ending time t p，end= t, calculate the current time t p，end The remaining useful life prediction result of Then enter step S4.7;

[0070] S4.7. Add the correction value to each of the currently obtained remaining useful life prediction values And remove the data with capacity regeneration phenomenon from the capacity degradation data X of the lithium battery to be tested Then let t = t p，begin + 1, and then return to step S4.1;

[0071] S5. When the capacity of the lithium battery to be tested degrades to the failure threshold w, the prediction is terminated and the algorithm ends.

[0072] In this embodiment, the LSTM neural network is used to perform real-time prediction of the RUL of the lithium battery B0005 at different times, and the results are as Figure 5 shown. It can be clearly seen that the RUL prediction curve obtained by the present invention has a small error from its actual RUL curve, and can provide accurate RUL information for the maintenance and support work of the lithium battery, thereby being beneficial to improving the reliability of the lithium battery during operation.

[0073] To quantitatively compare and measure the prediction performance, Figure 6 shows the RUL prediction results of the present invention, the non-linear Wiener process (the non-linear function uses at b , where a and b are parameters) and the long short-term memory neural network (LSTM) for the lithium battery B0005. Through Figure 6 the prediction results of different types of models for the RUL of the lithium battery, it can be found that since the present invention considers the capacity regeneration phenomenon in the degradation process of the lithium battery and corrects the error caused by the capacity regeneration phenomenon to the RUL prediction. Therefore, the present invention has a better effect than the traditional prediction model that does not consider the capacity regeneration phenomenon. At the same time, the capacity regeneration phenomenon detection algorithm proposed by the present invention can detect the capacity regeneration phenomenon during the degradation process of the lithium battery in real time, and can well adapt to the randomness of the capacity regeneration phenomenon, thereby providing more accurate RUL prediction results. Table 1 gives the average RUL prediction errors of each model for the lithium battery B0005.

[0074] LSTM Wiener process the present invention average error 30.8 cycle 10.2 cycle 1.6 cycle

[0075] Table 1

[0076] As can be seen from the prediction results shown in Table 1, the accuracy of the remaining useful life prediction results of this model is much higher than that of other models, which directly demonstrates the advantages of a new method for predicting the remaining useful life of a lithium battery considering the capacity regeneration phenomenon proposed by the present invention.

[0077] The above experimental results show that, compared with the existing data-driven lithium battery RUL prediction model, the new method for predicting the remaining life of lithium batteries considering the phenomenon of capacity regeneration proposed by the present invention has higher prediction accuracy and better real-time performance, and thus is more suitable for the needs of remaining life prediction in practical engineering.

[0078] Although the above-described illustrative specific embodiments of the present invention have been described to facilitate the understanding of the present invention by those skilled in the art, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are within the scope of protection.

Claims

1. A method for predicting the remaining life of a lithium battery considering the capacity regeneration phenomenon, characterized in that, it includes the following steps: (1). Obtain the capacity of the lithium battery at different times; By accelerating the life experiments of N lithium batteries and sampling the capacities of each lithium battery at different times, where the capacity of the i-th lithium battery at different times is denoted as where i = 1, 2, …, N, represents the initial capacity of the i-th lithium battery, represents the capacity of the i-th lithium battery at the t-th moment, t = 0, 1, 2, ... T, and T represents the number of sampling moments; (2). Detect the capacity regeneration phenomenon in the historical data of the lithium battery; (2.1) Obtain the first-order difference of the capacity data of the i-th lithium battery where (2.2) Obtain the first-order difference data of the first k that do not include the capacity regeneration phenomenon for the i-th lithium battery T > k > 2, and then calculate the mean μ i and the variance σ i : (2.3). Let k = k + 1, and judge whether k is less than T. If so, enter step (2.4); if not, enter step (3); (2.4), Determine whether it is greater than min(0, μ i + 3σ i ). If not, return to step (2.2); if so, proceed to step (2.5); (2.5) Set a counting variable q with an initial value of 0; let q = q + 1, and then record the starting time when the q-th capacity regeneration phenomenon occurs in the i-th lithium battery and the starting capacity corresponding to the starting time and proceed to step (2.6); (2.6) Again, let k = k + 1 and determine whether it is less than If not, re-enter (2.6); if so, record the end time when the qth capacity regeneration phenomenon occurs in the ith lithium battery and the end capacity corresponding to the end time and enter step (2.7); (2.7) Remove the capacity data X from the capacity data of the i-th lithium battery i Then let and then return to step (2.2); (3). Train the long short-term memory neural network LSTM; After the capacity data of N lithium batteries are processed through step (2), historical capacity data of lithium batteries without capacity regeneration phenomenon are obtained, denoted as where n is the number of data in each group; then an LSTM neural network is trained using these N groups of data; (4). Real-time prediction of the life of the lithium battery to be measured (4.1) Obtain the capacity degradation data X = [x 0 , x 1 ... x t of the lithium battery to be tested at the current moment, and determine whether x t is less than the failure threshold w. If it is less, go to step (5) and the algorithm ends; otherwise, go to step (4.2); (4.2), calculate the first-order difference ΔX of data X = [Δx 1 , Δx 2 ... Δx t , where Δx j = x j - x j-1 , j = 1, 2, …, t; then calculate the mean μ and variance σ; (4.3), Determine Δx t Is it greater than min(0, μ + 3σ)? If not, go to (4.4); if so, go to (4.5); (4.4) Input the data X = [x 0 , x 1 ... x t into the LSTM neural network to predict the remaining useful life RUL t of the lithium battery to be measured at time t. Then let t = t + 1, and then enter step (4.1); (4.5) Set a counting variable p with an initial value of 0; let p = p + 1, and then record the starting time t when the p-th capacity regeneration phenomenon occurs in the lithium battery to be measured p，begin = t - 1, and the starting capacity corresponding to the starting time Calculate the remaining life corresponding to the current time t as where represents the predicted remaining life value at time t obtained by the LSTM neural network, and then enter step (4.6); p，begin ​ (4.6) Let t = t + 1 again, and judge whether x t is less than If not, calculate the remaining life prediction value at the current moment as and re-enter (4.6); if so, record the end time t p，end when the p-th capacity regeneration phenomenon occurs in the lithium battery to be tested, calculate the remaining life prediction result at the current moment t p，end as Then enter step (4.7); (4.7) Add the correction value E = t p，end - t p，begin - 1 to each of the remaining life prediction values obtained currently, and remove the data with capacity regeneration phenomenon from the capacity degradation data X of the lithium battery to be measured Then let t = t p，begin + 1, and then return to step (4.1); (5). When the capacity of the lithium battery to be measured degrades to the failure threshold w, the prediction terminates and the algorithm ends.

Images