A lithium battery state of health estimation method based on dynamic working condition condition of variational mode decomposition

Abstract

This invention relates to a method for estimating the state of health (SOH) of lithium batteries under dynamic operating conditions based on variational mode decomposition (VMD). The method involves conducting lithium battery aging tests, performing dynamic operating condition cycles and standard operating condition discharge cycles on a reference lithium battery, and collecting time-series data on voltage, current, operating time, and temperature. The SOH sequence data and first-order difference SOH sequence data of the reference lithium battery are calculated using the ampere-hour integral method. Variational mode decomposition is then performed on the voltage, current, and temperature time-series data during the dynamic cycle operation. Furthermore, a fast Fourier transform (FFT) is applied to the decomposition results to extract health features from the frequency domain data. Finally, a long short-term memory (LSM) neural network is constructed and trained to achieve the estimation of the lithium battery's SOH under dynamic operating conditions. This method is more closely aligned with the actual operating conditions of lithium batteries and achieves accurate lithium battery SOH estimation.

Description

Technical Field

[0001] This invention relates to the field of battery health monitoring technology, and in particular to a method for estimating the health status of lithium batteries under dynamic operating conditions based on mode decomposition. Background Technology

[0002] Lithium-ion batteries, with their advantages of high energy density, long lifespan, and environmental friendliness, are considered the preferred technology for energy storage and supply in the new and renewable energy sectors. However, after undergoing charge-discharge cycles, the internal materials and physical properties of lithium-ion batteries change over time, leading to irreversible degradation in performance and lifespan. Compared to their initial state, aged batteries exhibit significantly reduced stored energy and output power, making it difficult to meet operational demands and potentially causing safety incidents. Therefore, timely and accurate estimation of the State of Health (SOH) is crucial for maintaining the efficient and safe operation of lithium-ion batteries.

[0003] The aging process of lithium batteries exhibits nonlinear characteristics, especially under dynamic charge-discharge conditions. The state of health (SOH) of lithium batteries is influenced by a combination of factors, including discharge rate, load conditions, temperature, and depth of discharge. Existing lithium battery SOH estimation methods primarily target ideal laboratory conditions with constant charge-discharge cycles. Their accuracy is poor under dynamic conditions, which more closely resemble real-world operating scenarios, making them unsuitable for practical applications. Therefore, this invention proposes a lithium battery SOH estimation method based on mode decomposition under dynamic operating conditions. This method extracts health features from time-series data of lithium batteries measured under dynamic operating conditions, thereby achieving SOH estimation. Summary of the Invention

[0004] To address the problem of inaccurate SOH estimation of lithium batteries under dynamic operating conditions in existing technologies, this invention provides a lithium battery health state estimation method based on mode decomposition under dynamic operating conditions, which is more consistent with the actual operating state of lithium batteries.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for estimating the state of health of a lithium battery under dynamic operating conditions based on mode decomposition includes the following steps:

[0007] Step 1: Conduct aging tests to obtain time-series data on voltage, current, operating time, and temperature during the reference lithium battery's dynamic operating conditions;

[0008] Step 2: After each w dynamic cycle, perform a standard operating condition discharge cycle on the reference lithium battery and collect timing data of voltage, current, running time and temperature.

[0009] Step 3: Calculate the discharge capacity of the reference lithium battery under standard operating conditions using the ampere-hour integration method to obtain the SOH sequence data and first-order differential SOH sequence data of the reference lithium battery;

[0010] Step 4: Perform variational mode decomposition on the time-series data of voltage, current and temperature of the reference lithium battery during w dynamic cycles;

[0011] Step 5: Perform time-frequency analysis on the modal decomposition results of each order in Step 4, and extract the maximum amplitude and corresponding frequency value from the frequency domain results as health features;

[0012] Step 6: Construct an initial long short-term memory neural network, and train it based on the health feature dataset in Step 5 and the first-order difference SOH sequence dataset in Step 3;

[0013] Step 7: Use the long short-term memory neural network model trained in step 6 to estimate the health status of the lithium battery under test operating under dynamic conditions.

[0014] Furthermore, the specific method of step 1 is as follows:

[0015] A reference lithium battery with similar specifications and operating conditions to the lithium battery under test was selected as the subject for aging tests. Under dynamic operating conditions, a discharge cycle was defined as the time it took for the reference lithium battery to discharge from its rated voltage to its termination voltage. Conversely, a charge cycle was defined as the time it took for the battery to charge from its termination voltage to its rated voltage. Each discharge cycle and charge cycle was also defined as one dynamic cycle.

[0016] During the entire lifespan of a reference lithium battery, the voltage data [V1, V2, ..., V] for n dynamic cycles are measured and recorded. n Current data [I1, I2, ..., I] n Running time data [t1, t2, ..., t] n ] and temperature data [T1, T2, ..., T n ].

[0017] Taking voltage datasets as an example, V n This represents the sequence of reference lithium battery voltage values ​​in the nth dynamic cycle, i.e. Where m n This represents the length of the time-series data collected in the nth dynamic cycle, and is related to the charge / discharge time. The meanings of the samples and subscripts in the current, time, and temperature data are the same as those in the voltage data.

[0018] Furthermore, the specific method for step 2 is as follows:

[0019] Each time the dynamic cycle in step 1 is run w times, a standard operating condition discharge cycle is performed on the reference lithium battery. The value of w depends on expert experience; the smaller the value, the higher the cost. A range of 30 to 50 cycles is generally recommended.

[0020] The definition and requirements of the standard operating condition discharge cycle for lithium batteries vary depending on different industry standards, and a unified definition has not yet been formed. This invention specifies that a standard operating condition discharge cycle is defined as the process of discharging a lithium battery from its rated voltage to its termination voltage at a 1C discharge current after constant current and constant voltage charging, followed by a 1-hour rest period at an ambient temperature of 23℃±2℃.

[0021] Over the entire lifespan of the reference battery, measure and record the voltage data [V1, V2, ..., V] for k standard operating condition discharge cycles. k Current data [I1, I2, ..., I] k Running time data [t1, t2, ..., t] k ] and temperature data [T1, T2, ..., T k Over its entire lifespan, the number of dynamic cycles of a lithium battery satisfies the relationship n = w × k.

[0022] Taking voltage datasets as an example, V k This represents the sequence of voltage values ​​of the reference lithium battery during the k-th standard operating condition discharge cycle, i.e. Among them o k The length of the time-series data collected during the k-th standard operating condition discharge cycle is related to the discharge time. The meanings of the samples and subscripts in the current, time, and temperature data are the same as those in the voltage data.

[0023] Furthermore, the specific method of step 3 is as follows:

[0024] The health status of a lithium battery can be characterized by its capacity, expressed as the ratio between the discharge capacity under current aging conditions and the initial rated capacity. The expression is as follows:

[0025]

[0026] Among them, C batt C represents the discharge capacity of a lithium battery under its current aging state. init This indicates the initial rated capacity of the lithium battery.

[0027] The discharge capacity of a reference lithium battery during the kth standard operating condition discharge cycle is calculated using the ampere-hour integral method, and its expression is as follows:

[0028] C k =∫0 t idτ

[0029] Where i is the discharge current under standard operating conditions, t is the discharge time under standard operating conditions, and C k This represents the discharge capacity value.

[0030] For a reference lithium battery undergoing k standard operating condition discharge cycles, the calculated SOH sequence data is SOH=[SOH1, SOH2, ..., SOH... k Furthermore, the first-order difference SOH sequence data of the reference lithium battery can be calculated as ΔSOH=[ΔSOH1, ΔSOH2, ..., ΔSOH]. k The formula for calculating ]′ is as follows:

[0031]

[0032] Wherein, ΔSOH i For the i-th difference SOH result, SOH i and SOH i-1 These are the reference lithium battery health status values ​​for the i-th and i-1th cycles, respectively.

[0033] Furthermore, the specific method of step 4 is as follows:

[0034] By concatenating the time-series data of voltage, current, and temperature collected from the reference lithium battery during w dynamic cycles, we can obtain k sets of time-domain datasets:

[0035]

[0036] For the voltage, current, and temperature data in the k sets of time-domain datasets, ω-order variational mode decomposition (VMD) is performed respectively, and the corresponding mode function results are as follows:

[0037]

[0038] The order ω of the VMD method can be set based on experience, taking into account both decomposition accuracy and computational complexity. The recommended range is usually 4 to 6.

[0039] Furthermore, the specific method of step 5 is as follows:

[0040] The Fast Fourier Transform (FFT) was used to perform time-frequency analysis on the mode decomposition results obtained in step 4. The voltage, current, and temperature time-domain signals of the reference lithium battery dynamic cycling process were transformed to the frequency domain to obtain the corresponding k sets of spectral information of the dynamic cycling process:

[0041]

[0042] Furthermore, the maximum amplitude and its corresponding frequency value are extracted from the ω-order spectral data of voltage, current, and temperature, respectively, as health features, forming the health feature dataset:

[0043]

[0044] Furthermore, the specific method for step 6 is as follows:

[0045] Initialize the Long Short-Term Memory (LSTM) neural network, using the health feature dataset from step 5 as input data and the first-order difference SOH sequence dataset from step 3 as health status label data. Define the loss function of the LSTM neural network as follows:

[0046]

[0047] Among them, Y o This represents the training output value of the LSTM neural network, and SOH represents the corresponding true label value.

[0048] Furthermore, the input data and label data of the reference lithium battery are divided into training and test sets, and the Adam algorithm is used for gradient descent to finally obtain the trained LSTM neural network model.

[0049] Furthermore, the specific method for step 7 is as follows:

[0050] During the dynamic operation of the lithium battery under test, the time-series data of voltage, current, and temperature in each dynamic cycle are measured and recorded for ω-order variational mode decomposition. Step 5 is repeated, and the ω-order variational mode decomposition results are subjected to fast Fourier transform. At the same time, the maximum amplitude and corresponding frequency values ​​are extracted as health feature data and input into the LSTM neural network model trained in step 6 to calculate the first-order difference SOH result, thereby realizing the health state estimation of the lithium battery under test.

[0051] The advantages of this invention are: it considers the real operating scenarios of lithium batteries, estimates the health status of lithium batteries under dynamic operating conditions, extracts health features from the data collected during dynamic operating conditions, and thus assesses the capacity degradation status of lithium batteries. It has the advantages of high computational efficiency and high time-series prediction accuracy. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the implementation steps of the present invention.

[0053] Figure 2This is a voltage curve of a reference lithium battery under standard operating conditions during discharge, as shown in this embodiment of the invention.

[0054] Figure 3 This is a SOH degradation curve of a reference lithium battery in an embodiment of the present invention.

[0055] Figure 4 Reference lithium batteries V1 to V in the embodiments of the present invention 40 VMD decomposition results of time series data

[0056] Figure 5 Reference lithium batteries V1 to V in the embodiments of the present invention 40 Time series data FFT result graph

[0057] Figure 6 This is a graph showing the SOH estimation results of the lithium battery under test in an embodiment of the present invention.

[0058] Figure 7 This is a graph showing the RMSE error of the SOH estimation results of the lithium battery under test in an embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0060] like Figure 1 The diagram shown is a flowchart illustrating a method for estimating the health status of a lithium battery under dynamic operating conditions based on mode decomposition, according to an embodiment of the present invention. Figure 1 The method shown includes the following steps:

[0061] Step 1: Conduct aging tests to obtain time-series data on voltage, current, operating time, and temperature during the reference lithium battery's dynamic operating conditions;

[0062] In this embodiment of the invention, two commercially available cylindrical 18650 lithium batteries of the same model and batch, with a rated capacity of 2Ah, were selected. One of them was selected as a reference lithium battery for aging tests. During the discharge cycle, a discharge current between 0.5A and 4A was randomly selected, with each discharge cycle lasting 5 minutes, until the cutoff voltage of 3.2V was reached. During the charging cycle, a constant current of 2A was maintained to charge to the rated voltage of 4.2V, and then the current was switched to constant voltage mode until the current dropped to 20mA.

[0063] The above charging and discharging process is recorded as one dynamic cycle. Referring to a lithium battery undergoing 840 dynamic cycles, voltage data [V1, V2, ..., V] are measured and obtained. 840 Current data [I1, I2, ..., I] 840 ] , runtime data [T1, T2, ..., T 840 ] and temperature data [T1, T2, ..., T 840 ].

[0064] Step 2: After each w dynamic cycle, perform a standard operating condition discharge cycle on the reference lithium battery and collect timing data of voltage, current, running time and temperature.

[0065] Each time the dynamic cycle in step 1 is run w times, a standard operating condition discharge cycle is performed. This invention specifies that after the lithium battery undergoes constant current and constant voltage charging, it is left to stand for 1 hour. The discharge process, characterized by a 1C discharge current decreasing from the battery's rated voltage to the termination voltage at an ambient temperature of 23℃±2℃, is considered a standard operating condition discharge cycle.

[0066] In this embodiment of the invention, w = 40, meaning that a standard operating condition discharge cycle is performed every 40 dynamic cycles. Over the entire lifespan of the reference lithium battery, a total of 21 standard operating condition discharge cycles are performed, and the voltage curve is shown below. Figure 2 As shown. Voltage data [V1, V2, ..., V] were obtained under standard operating conditions. 21 Current data [I1, I2, ..., I] 21 Running time data [t1, t2, ..., t] 21 ] and temperature data [T1, T2, ..., T 21 ].

[0067] Step 3: Calculate the discharge capacity of the reference lithium battery in standard operating condition discharge cycles using the ampere-hour integration method to obtain the SOH sequence dataset and the first-order difference SOH sequence dataset of the reference lithium battery;

[0068] The health status of a lithium battery can be characterized by its capacity, expressed as the ratio between the discharge capacity under current aging conditions and the initial rated capacity. The expression is as follows:

[0069]

[0070] Among them, C batt C represents the discharge capacity of a lithium battery under its current aging state. init This indicates the initial rated capacity of the lithium battery.

[0071] The discharge capacity of a reference lithium battery during the kth standard operating condition discharge cycle is calculated using the ampere-hour integral method, and its expression is as follows:

[0072] C=∫0 t idτ

[0073] Where i is the discharge current under standard operating conditions, t is the discharge time under standard operating conditions, and C k This represents the discharge capacity value.

[0074] In this embodiment of the invention, the healthy state state (SOH) sequence data SOH=[1.94324,1.86394,1.79934,1.75317,...,1.06199]' was calculated based on 21 standard operating condition discharge cycles of a lithium battery, and its degradation curve is shown below. Figure 3 As shown. Further, solve the first-order difference SOH sequence dataset ΔSOH=[-0.06270,-0.07929,-0.06459,-0.04617,...,-0.03383]'.

[0075] Step 4: Perform variational mode decomposition on the time-series data of voltage, current and temperature obtained from the reference lithium battery during w dynamic cycles;

[0076] In this embodiment of the invention, for 840 dynamic cycles of a reference lithium battery, the time-series data of voltage, current, and temperature collected every 40 dynamic cycles are concatenated into a dataset, resulting in 21 sets of time-domain datasets:

[0077]

[0078] Based on the aforementioned time-domain dataset, a 5th-order variational mode decomposition (VMD) was performed on the voltage, current, and temperature time-series data of the reference lithium battery, yielding the corresponding mode function results:

[0079]

[0080] Using time series data [V1, V2, ..., V 40 For example, its original data and the 5th-order VMD results obtained through variational mode decomposition are as follows: Figure 4 As shown.

[0081] Step 5: Perform time-frequency analysis on the modal decomposition results of each order in Step 4, and extract the maximum amplitude and corresponding frequency value from the frequency domain results as health features;

[0082] In this embodiment of the invention, the 5th-order VMD decomposition results corresponding to the original time-domain data of voltage, current, and temperature of the reference lithium battery dynamic cycling process in step 4 are transformed to the frequency domain space, resulting in 315 sets of frequency domain data of VMD for 21 sets of dynamic cycling processes:

[0083]

[0084] Using time series data [V1, V2, ..., V 40 For example, the 5th-order VMD decomposition result is subjected to a Fast Fourier Transform (FFT) to obtain the spectral data [f]. V11 f V12 , ..., f V15 The result is as follows Figure 5 As shown.

[0085] Furthermore, the maximum amplitude and its corresponding frequency value were extracted from the fifth-order spectral data of voltage, current, and temperature to form the health feature dataset:

[0086]

[0087] Step 6: Construct an initial long short-term memory neural network, and train it based on the health feature dataset in Step 5 and the first-order difference SOH sequence dataset in Step 3;

[0088] In this embodiment, the health feature dataset from step 5 is used as the input data, and the first-order difference SOH sequence dataset from step 3 is used as the output data. The loss function for constructing the Long Short-Term Memory (LSTM) neural network is as follows:

[0089]

[0090] Among them, Y o This represents the training output value of the LSTM neural network, and SOH represents the corresponding true label value.

[0091] Furthermore, the input data and label data of the reference lithium battery are divided into training and test sets, and the Adam algorithm is used for gradient descent to finally obtain the trained LSTM neural network model.

[0092] Step 7: Use the long short-term memory neural network model trained in step 6 to estimate the health status of the lithium battery under test.

[0093] During the dynamic operation of the lithium battery under test, a fifth-order variational mode decomposition was performed on the time-series data of voltage, current, and temperature acquired in each dynamic cycle. Step 5 was repeated, and the time-domain results of voltage, current, and temperature in each dynamic cycle were subjected to a fast Fourier transform for time-frequency analysis.

[0094] Furthermore, the maximum amplitude and corresponding frequency values ​​are extracted as a health feature dataset and input into the LSTM neural network model trained in step 6 to obtain the first-order difference SOH result, thereby achieving the health status estimation of the lithium battery under test. In this embodiment, the SOH estimation result and RMSE error of the lithium battery under test are as follows: Figure 6 and Figure 7 As shown.

[0095] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A method for lithium battery state of health estimation under dynamic operating conditions based on variational modality decomposition, characterized in that: The method comprises the following steps: Step 1: carry out aging test, obtain the time sequence data of voltage, current, running time and temperature of the reference lithium battery during dynamic working condition operation; Step 2: Whenever running After the sub-dynamic cycle, a standard working condition discharge cycle is carried out on the reference lithium battery, and the time series data of voltage, current, running time and temperature are collected; Step 3: calculate the dischargeable capacity of the reference lithium battery under the standard working condition discharge cycle by using the ampere-hour integral method, obtain the SOH sequence data and the first-order difference SOH sequence data of the reference lithium battery; Step 4: The time series data of voltage, current and temperature of the reference lithium battery in the sub-dynamic cycle process are decomposed by variational mode decomposition; Step 4: The time series data of voltage, current and temperature of the reference lithium battery in the sub-dynamic cycle process are decomposed by variational mode decomposition; Step 5: carry out time-frequency analysis on the results of each order modal decomposition in step 4, and extract the maximum amplitude and corresponding frequency value from the frequency domain result as the health characteristics; Step 6: construct an initialized long short-term memory neural network, and train the long short-term memory neural network based on the health characteristic data set in step 5 and the first-order difference SOH sequence data set in step 3; Step 7: use the long short-term memory neural network model trained in step 6 to estimate the health state of the lithium battery to be tested running in dynamic working condition; Through the above steps, a lithium battery health state estimation method under dynamic working condition based on variational modal decomposition is given.

2. The lithium battery health state estimation method under dynamic working condition based on variational modal decomposition according to claim 1, wherein: measuring and recording, over the full life cycle of a reference lithium battery voltage data , current data , runtime data and temperature data ; Taking voltage datasets as an example, Indicates the first The voltage sequence of the reference lithium battery during the next dynamic cycle, i.e. ,in Representing the The length of the time-series data collected in each dynamic cycle is related to the charging and discharging time; the meaning of the samples and subscripts in the current data, time data, and temperature data is the same as that in the voltage data.

3. The lithium battery health state estimation method under dynamic working condition based on variational modal decomposition according to claim 1, wherein: In step 2, whenever running In the dynamic cycle in substep 1, the reference lithium battery is discharged once under a standard working condition, wherein, The value depends on the experience of an expert, and the smaller the value, the higher the cost consumed, and the range is 30-50 times. measuring and recording voltage data , current data , run time data and temperature data of the secondary standard discharge cycles over the life of the reference battery; and the number of dynamic cycles of the lithium battery over the life satisfies the relationship Taking voltage datasets as an example, Indicates the first The voltage sequence of the reference lithium battery during sub-standard discharge cycles, i.e. ,in Representing the The length of the time-series data collected during a standard operating condition discharge cycle is related to the discharge time. The meanings of the samples and subscripts in the current data, time data, and temperature data are the same as those in the voltage data.

4. The lithium battery health state estimation method under dynamic working condition based on variational modal decomposition according to claim 1, wherein: For the reference lithium battery The SOH sequence data calculated for the sub-standard operating condition discharge cycle process is Further, the first-order differential SOH sequence data of the reference lithium battery can be calculated as The calculation formula is as follows: wherein, is the th difference SOH result, and are the th and th reference lithium battery state of health values, respectively.

5. The lithium battery health state estimation method under dynamic working condition based on variational modal decomposition according to claim 1, wherein: In step 4, the reference lithium battery is used. The time-series data of voltage, current, and temperature collected during each dynamic cycle are concatenated into a dataset, which can yield... The group time-domain dataset is as follows: For The voltage, current and temperature data in the group time domain data set are respectively processed by The corresponding modal function results obtained by the order variable mode decomposition are: The order of the variational modality decomposition method Based on the experience setting, considering the decomposition accuracy and the calculation complexity, the range is: 4~6.

6. The lithium battery health state estimation method under dynamic working condition based on variational modal decomposition according to claim 1, wherein: In step 5, the fast Fourier transform is used to perform time-frequency analysis on the modal decomposition results obtained in step 4, and the voltage, current and temperature time domain signals of the reference lithium battery dynamic cycle process are transformed into the frequency domain space to obtain the corresponding The frequency spectrum information of the group dynamic cycle process is: Further, the maximum amplitude and its corresponding frequency value in the order spectrum data of the voltage, the current and the temperature are extracted as health features, respectively, to form a health feature dataset as follows: Further, the maximum amplitude and its corresponding frequency value in the order spectrum data of the voltage, the current and the temperature are extracted as health features, respectively, to form a health feature dataset as follows: 。 7. The lithium battery health state estimation method under dynamic working condition based on variational modal decomposition according to claim 1, wherein: In step 6, the long short-term memory neural network is initialized, the health characteristic data set in step 5 is used as input data, the first-order difference SOH sequence data set in step 3 is used as health state label data, and the loss function of the long short-term memory neural network is defined as: wherein, denotes a training output value of a long short-term memory neural network, denotes a corresponding true label value; Further, the input data and label data of the reference lithium battery are divided into training set and test set, Adam algorithm is used for gradient descent, and finally the trained long short-term memory neural network model is obtained.

8. The lithium battery health state estimation method under dynamic working condition based on variational modal decomposition according to claim 1, wherein: In step 7, during the dynamic operation of the lithium battery under test, the time-series data of voltage, current, and temperature in each dynamic cycle are measured and recorded. Perform variational mode decomposition, repeat step 5, and... The first-order variational mode decomposition results are subjected to fast Fourier transform, and the maximum amplitude and corresponding frequency values ​​are extracted as health feature data. These data are then input into the long-term time memory neural network model trained in step 6 to calculate the first-order difference SOH result, thereby achieving the health status estimation of the lithium battery under test.

Images