CN-120629954-B - Method for predicting service life of lithium ion battery based on early-stage circulation data

Abstract

The invention discloses a method for predicting the service life of a lithium ion battery based on early-stage circulation data, and belongs to the technical field of power battery management. According to the method, a Q-U curve is generated by collecting discharge voltage and capacity data of the previous 100 charge-discharge cycles of the battery, a Q-U curve collection point is aligned through a Gaussian process regression method, a Life Factor Vector (LFV) is extracted after average treatment, the Q-U curve collection point is used as an input characteristic of a gate-controlled cyclic neural network (GRU), the cycle number when the battery capacity is attenuated to 80% is used as an output target, and a life prediction model is trained. And optimizing super parameters by adopting five-fold cross validation, so as to ensure the model precision. For the battery to be tested, the cycle life of the battery can be predicted by only performing the first 100 cycles. Experiments show that the method can predict the service life of batteries with the same capacity and different charge and discharge cycles, has the relative error as low as 3.68 percent, has higher accuracy and engineering practicability, and improves the safety of a battery system.

Inventors

• ZHANG YANQIN
• YANG ZEHAO

Assignees

• 北京工业大学

Dates

Publication Date

20260512

Application Date

20250604

Claims (2)

1. 1. A method for predicting the service life of a lithium ion battery based on early cycle data is characterized in that the method comprises the steps of extracting a Q-U curve of a discharge interval of the previous 100 cycles of the lithium ion battery, uniformly aligning acquisition points of the Q-U curve through Gaussian process regression, calculating an average Q-U curve and a service life factor vector LFV, establishing a prediction model by utilizing a GRU neural network, optimizing super parameters by adopting five-fold cross validation to obtain a final trained prediction model; The specific implementation steps are as follows: Step 1, arranging charge-discharge cycle data of the existing battery, wherein batteries with the same positive and negative electrode materials and the same capacity are required, and the discharge cycle should completely record the data of discharge voltage, current and capacity changing along with time; step 2, extracting discharge voltage of the 6 th to 10 th and 96 th to 100 th cycles and corresponding accumulated discharge capacity data from the first 100 cycles of data of each battery to obtain a Q-U curve, wherein the abscissa is the discharge voltage U of the battery, and the ordinate is the accumulated discharge capacity Q discharged from a full charge state to the corresponding voltage; step 3, ensuring that the number and the numerical value of the voltage sampling points of each extracted Q-U curve are the same; and (4) calculating and obtaining a life factor vector LFV by adopting the formula (4): (4); In the formula, 、 The capacity data of each acquisition point of the Q (U) l curve is subtracted from the capacity data of each corresponding acquisition point of the Q (U) e curve to obtain LFV, wherein the Q values correspond to each voltage point on the Q (U) e and the Q (U) l respectively; Step 5, inputting a life factor LFV of each training battery as a prediction model, recording the cycle number which is carried out at the moment as the battery cycle life when the capacity of the battery is reduced to 80% of the initial capacity, and outputting the cycle number as the prediction model; Step 6, optimizing the super parameters of the prediction model by adopting a five-fold cross validation method, equally dividing the training set data into 5 subsets, selecting 4 subsets of the training set data as training data each time, selecting the remaining 1 subsets as validation data, repeating each group of super parameters for 5 times to ensure that each subset is used for validation once, finally taking the average error of the 5 times of validation as an evaluation index of the influence of the group of super parameters on the performance of the prediction model, and judging the result of super parameter optimization by using Root Mean Square Percentage Error (RMSPE), wherein the result is shown in a formula (5): (5); In the middle of In order to predict the number of samples of the result, As the life prediction value of the i-th battery, When the error of the RMSPE obtained after the cross validation of a group of super parameter combinations is lower than 3%, selecting the group of super parameters and finishing the prediction model training; And 7, when the service life of the target battery is predicted, only 100 times of cyclic charge and discharge tests are needed to be carried out on the battery, and the service life factor LFV is extracted according to the steps 2 to 4 and then is input into a trained prediction model, wherein the prediction model outputs the predicted service life of the target battery.
2. 2. The method for predicting lithium ion battery life based on early cycle data of claim 1, wherein the specific steps of step (3) are as follows: 1) Establishing a Gaussian process regression model GPR for each Q-U curve, taking a discharge voltage U as an input, taking 6 th to 10 th and 96 th to 100 th cycles, accumulating a discharge capacity Q as an output, modeling a nonlinear relation by using a square index kernel function, and training the Gaussian regression model: (1); Wherein the method comprises the steps of For the signal variance to be a function of the signal variance, In the form of a length scale, As the variance of the noise is the value of the variance of the noise, Is a Kronecker delta function; automatic optimization of superparameters by maximum edge likelihood : (2); The optimization process is concretely to initialize the super parameters And calculates covariance matrix K, and iteratively updates using gradient-increasing method Until the edge likelihood converges; 2) Designating a voltage variation range according to the minimum value and the maximum value of the voltage variation in the discharging process, and equally dividing the voltage range into 1000 equidistant points Predicting each using the trained prediction model Corresponding to the point Values, each Q-U curve yields a resampled aligned curve as shown in equation (3): (3) ; And (3) for each battery, after resampling and aligning 10Q-U curves, respectively calculating the average value of Q corresponding to each U acquisition point on the Q-U curves of the 6 th to 10 th cycles and the 96 th to 100 th cycles, and obtaining two Q-U curves Q (U) e and Q (U) l after averaging.

Description

Method for predicting service life of lithium ion battery based on early-stage circulation data Technical Field The invention belongs to the technical field of power battery management, and is used for predicting the cycle life of a vehicle-mounted lithium ion battery under the condition of rapid charge and discharge. Background The lithium ion battery is widely applied to portable electronic equipment, electric automobiles and energy storage systems by virtue of the capabilities of high energy density, long service life, low self-discharge power, rapid charge and discharge and the like, and provides a clean and efficient energy storage scheme. The State of health (SOH) of a lithium ion battery reflects the ratio of the current performance of the battery to the initial performance and is an important parameter for representing the State of the battery. Depending on the application, a battery is considered near or at the end of its life when its SOH falls to 80% to 60% of its initial value. Degradation of battery performance not only affects device performance, but may also present a potential safety hazard. Therefore, if the cycle life of the battery can be accurately predicted when the battery is in early cycle, the use and maintenance of the battery can be planned in advance, a reasonable replacement plan can be formulated, potential safety risks are reduced, and the method has important significance in improving the reliability and safety of the system. Disclosure of Invention The invention provides a prediction method based on early cycle data in order to achieve the aim of predicting the cycle life of a lithium ion battery. The specific invention comprises the following steps of selecting the cycle life of each battery from the existing charge-discharge cycle data of the batch batteries as a target value (output), extracting life factor vectors (LFV, life Feature Vector) from the early cycle data (the previous 100 charge-discharge cycles are taken as an example) of each battery as input, establishing a data model by using a gate-controlled cyclic neural network (GRU, gated Recurrent Unit), using the LFV of the batch batteries as input, using the cycle life as an output training model, and optimizing the super parameters by using a cross-validation method to obtain the data model based on the GRU. For a predicted target battery, early cycle data, such as 100 cycles of data, are needed, life factor vectors are extracted from the cycle data, and the life factor vectors are input into a data model to predict the cycle life of the battery. Through practice, the life prediction is carried out on 10 batteries with different charge-discharge cycle modes, the relative error of the predicted life is as low as 3.68%, and the method can provide a powerful tool for health management and life assessment of batteries with the same capacity and different charge-discharge cycles. The complete technical route of the present invention is shown in fig. 1. The following describes the specific implementation steps: and step 1, arranging charge and discharge cycle data of the existing battery, wherein batteries with the same positive and negative electrode materials and the same capacity are required, and the discharge cycle should completely record the data of discharge voltage, current and capacity changing along with time. And 2, acquiring a Q-U curve. And extracting the discharge voltage of the 6 th to 10 th cycles and the 96 th to 100 th cycles and corresponding accumulated discharge capacity data from the earlier 100 th cycle data of each battery to obtain 10Q-U curves of each battery, wherein the abscissa is the discharge voltage U (V) of the battery, and the ordinate is the accumulated discharge capacity Q (Ah) from full charge state to corresponding voltage. Step 3:Q-resampling alignment and averaging calculation of U-curve data. In order to effectively reduce errors caused by capacity rise, an average Q-U curve is required to be calculated, meanwhile, in order to calculate LFV, 10 voltage sampling points of the Q-U curves are required to be unified, so that the voltage sampling points of each Q-U curve are consistent, and each voltage sampling point has a corresponding Q value. The Gaussian process regression system is adopted to align the voltage sampling points of each Q-U curve, and the specific steps are as follows: 1) A Gaussian process regression model (GPR) was built for each Q-U curve (cycles 6-10 and 96-100) with discharge voltage U as input and accumulated discharge capacity Q as output. Modeling a nonlinear relation by using a square index kernel (SE kernel) to train a Gaussian regression model: Wherein the method comprises the steps of Is the signal variance, l is the length scale,For noise variance, δ ij is the Kronecker delta function. Automatic optimization of superparameters by maximum edge likelihood The optimization process is specifically to initialize the super-parameter theta and calculate the