A Lithium Battery Health State Estimation Method Based on Local Capacity Increment Features

Abstract

The present invention discloses a method for estimating the health state of a lithium battery based on local capacity increment features, including: obtaining the charging data of the lithium battery and evaluating the voltage range corresponding to the state of charge from 20% to 80%; dividing each round of voltage range into voltage intervals and performing voltage repair on the voltage intervals; dividing each voltage interval after voltage repair into voltage sub-intervals and calculating the corresponding local voltage capacity increment for each sub-interval; inputting the local voltage capacity increment into a support vector regression model for training until the root mean square error loss function converges to obtain an optimized voltage interval; collecting the charging segment of the lithium battery at the current charge-discharge cycle number in real time, selecting the corresponding optimized voltage interval according to the charging segment to obtain a joint estimation value of multiple voltage intervals, and obtaining the optimal estimation of the current health state of the lithium battery through the Kalman filtering algorithm. The present invention solves the problem of difficult health state estimation of the data-driven model of the unmanned transport vehicle.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium batteries, and specifically relates to a method for estimating the state of health of a lithium battery based on local capacity increment characteristics. Background Art

[0002] With the improvement of the automation level in labor-intensive handling places such as factories and warehouses, automated guided vehicles are gradually being widely popularized. Due to advantages such as high energy density, long life, and low discharge rate, lithium-ion power batteries have gradually replaced the lead-acid batteries used in automated guided vehicles, but their usage risks are higher than those of lead-acid batteries. Therefore, how to timely estimate the state of health of the battery, evaluate and process units with unqualified life is of great significance for ensuring operation safety and avoiding losses.

[0003] Taking the use of lithium-ion batteries in electric vehicles as the background, state-of-health estimation algorithms based on data-driven models have been widely studied. However, different from the usage characteristics of electric vehicles, the charge and discharge cycles of automated guided vehicles are more frequent and the depth of discharge is lower, and previous state-of-health estimation algorithms often require data with a long information span, so it is difficult to be used in the scenario of automated guided vehicles.

[0004] For the usage scenario of automated guided vehicles, the present invention proposes a method for estimating the state of health of a lithium battery based on local capacity increment characteristics, which aims to fully and accurately complete the state-of-health estimation of automated guided vehicles while applying the characteristics of low depth of discharge. Summary of the Invention

[0005] Aiming at the above deficiencies in the prior art, the present invention provides a method for estimating the state of health of a lithium battery based on local capacity increment characteristics, which solves the problem of difficult state-of-health estimation of data-driven models for automated guided vehicles.

[0006] In order to achieve the above invention purpose, the technical solution adopted by the present invention is: a method for estimating the state of health of a lithium battery based on local capacity increment characteristics, specifically including the following steps:

[0007] S1. Obtain the charging data of each charge and discharge cycle during the process of the lithium battery from factory to retirement. Taking the first charge in the charge and discharge cycle as the standard, evaluate the voltage range corresponding to the state of charge from 20% to 80%, and retain all the charging data within the voltage range in the charging data;

[0008] S2. Divide the voltage range in the retained charging data of each round into N 1 voltage intervals at an interval of ΔU 1 and perform voltage repair on the N 1 voltage intervals;

[0009] S3. Divide each voltage interval for voltage repair into N voltage sub - intervals at intervals of ΔU, and calculate the local voltage capacity increment corresponding to each sub - interval; 2 divide into N 2 voltage sub - intervals, and calculate the local voltage capacity increment corresponding to each sub - interval;

[0010] S4. Input the local voltage capacity increment corresponding to each sub - interval into the support vector regression model for training until the mean square root error loss function converges, and obtain N lithium - ion battery state of health estimation models; 1 lithium - ion battery state of health estimation models;

[0011] S5. Real - time collect the lithium - battery charging segment at the current charge - discharge cycle number, select the corresponding lithium - ion battery state of health estimation model according to the charging segment, obtain the multi - model joint estimation value, and obtain the optimal estimation of the current lithium - battery state of health through the Kalman filter algorithm.

[0012] Further, the interval ΔU 1 is 50 mV - 200 mV.

[0013] Further, the interval ΔU 2 is 3 mV - 10 mV.

[0014] Further, the process of voltage repair for N voltage intervals is as follows: the charging voltage curve increases monotonically with the state of charge. If the voltage value at any moment during charging is lower than all previous moments, the voltage value at this moment is abnormal. Find the normal voltages at the two adjacent moments on the left and right of this abnormal voltage value, and use linear interpolation to replace the abnormal voltage value: 1 The charging voltage curve increases monotonically with the state of charge. If the voltage value at any moment during charging is lower than all previous moments, the voltage value at this moment is abnormal. Find the normal voltages at the two adjacent moments on the left and right of this abnormal voltage value, and use linear interpolation to replace the abnormal voltage value:

[0015]

[0016] where t abnormal is the moment when the abnormal voltage appears, is the moment adjacent to the left of the abnormal voltage, is the moment adjacent to the right of the abnormal voltage, is the corresponding normal voltage value, is the corresponding normal voltage value, V renew is the repaired value of the abnormal voltage.

[0017] Further, the local capacity increment Δq corresponding to the sub - interval is:

[0018]

[0019] where i(t) is the current corresponding to the lithium - battery at time t during charging, t st is the starting moment of the sub - interval, t edis the termination time of the sub-interval.

[0020] Further, the root mean square error loss function δ a is:

[0021]

[0022] where M val is the number of charging data samples in the a-th voltage interval, u is the index of M val , y val (u) represents the true value of the state of health of the battery for the u-th charging data sample, represents the estimated value of the state of health of the battery for the u-th charging data sample in the a-th voltage interval.

[0023] Further, step S5 includes the following sub-steps:

[0024] S5.1. Real-time collect the lithium battery charging segment at the current charge-discharge cycle number, select the corresponding lithium-ion battery state of health estimation model according to the charging segment, calculate the root mean square error of each lithium-ion battery state of health estimation model, and set the weight w of each lithium-ion battery state of health estimation model j :

[0025]

[0026] where J is the number of lithium-ion battery state of health estimation models corresponding to the charging segment, j represents the index of J, and δ j is the root mean square error between the j-th lithium-ion battery state of health estimation model and the real-time lithium battery charging segment;

[0027] S5.2. Estimate the state of health of the lithium battery through the corresponding lithium-ion battery state of health estimation model and combine the weights of each lithium-ion battery state of health estimation model to obtain a multi-model joint estimation value

[0028]

[0029] where is the matrix set composed of the state of health of the lithium battery estimated by each lithium-ion battery state of health estimation model, and W is the weight set of the lithium-ion battery state of health estimation model;

[0030] S5.3. Further correct the multi-model joint estimation value using the Kalman filter to obtain the optimal estimation of the current lithium battery state of health estimation

[0031] where is the prior estimate of the current charge-discharge cycle number k, and K(k) is the Kalman gain of the current charge-discharge cycle number k. P - (k) is the prior estimate variance of the current charge-discharge cycle number k, and P - (k) = P(k - 1) + Q, where P(k - 1) is the optimal estimate variance at the k - 1 charge-discharge cycle number, Q is the process variance, and R(k) is the estimate variance of the current charge-discharge cycle number k.

[0032] Compared with the prior art, the present invention has the following beneficial effects: The method for estimating the state of health of a lithium battery based on the local capacity increment feature of the present invention obtains multiple optimized voltage intervals through dividing voltage intervals and training, reduces the search for long-span information in each voltage interval, and greatly improves the probability of the battery management system achieving the estimation of the state of health of the battery. The method for estimating the state of health of a lithium battery of the present invention realizes a relatively accurate estimation of the state of health of the battery through the joint estimation of multiple voltage intervals, and further introduces the historical estimation result through Kalman filtering and dynamically adjusts its estimation variance to correct the joint estimation result, thereby improving the accuracy of the estimation of the state of health of the lithium battery once again. Description of the Drawings

[0033] Figure 1 is the flowchart of the method for estimating the state of health of a lithium battery based on the local capacity increment feature of the present invention. Detailed Embodiments

[0034] The following describes the detailed embodiments of the present invention to facilitate those skilled in the art to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the detailed embodiments. For those of ordinary skill 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.

[0035] As Figure 1 is the flowchart of the method for estimating the state of health of a lithium battery based on the local capacity increment feature of the present invention, and the method for estimating the state of health of the lithium battery specifically includes the following steps:

[0036] S1. Obtain the charging data of each charge-discharge cycle during the process of the lithium battery from factory to retirement. Taking the first charge in the charge-discharge cycle as the standard, evaluate the voltage range corresponding to the 20% to 80% state of charge, and retain all the charging data within this voltage range in the charging data.

[0037] S2. Divide the voltage range in the retained charging data of each round into N 1 voltage intervals at an interval of ΔU 1 , and for N 1Voltage repair is performed in a voltage interval to ensure the monotonicity of the charging voltage. The interval ΔU in the present invention 1 is set to 50 mV - 200 mV, so that the time corresponding to each voltage interval in the N 1 voltage intervals is not too long, thereby reducing the difficulty of collecting features in actual use.

[0038] In the present invention, the process of voltage repair for the N 1 voltage intervals is as follows: The charging voltage curve increases monotonically with the state of charge. If the voltage value at any moment during charging is lower than all previous moments, the voltage value at that moment is abnormal. Find the normal voltages at the two adjacent moments on the left and right of the abnormal voltage value, and replace the abnormal voltage value by linear interpolation:

[0039]

[0040] where t abnormal is the moment when the abnormal voltage appears, is the moment adjacent to the left of the abnormal voltage, is the moment adjacent to the right of the abnormal voltage, is for the corresponding normal voltage value, is for the corresponding normal voltage value, V renew is the repaired value of the abnormal voltage.

[0041] S3. Divide each voltage interval after voltage repair into N 2 voltage sub-intervals according to the interval ΔU 2 and calculate the local voltage capacity increment corresponding to each sub-interval; the interval ΔU in the present invention 2 is set to 3 mV - 10 mV, and the local voltage capacity increment obtained at this interval can well reflect the aging process of the lithium battery.

[0042] The local capacity increment Δq corresponding to the sub-interval in the present invention is:

[0043]

[0044] where i(t) is the current corresponding to the lithium battery at time t during charging, t st is the start time of the sub-interval, and t ed is the end time of the sub-interval.

[0045] S4. Input the local voltage capacity increment corresponding to each sub-interval into the support vector regression model for training until the mean square root error loss function converges, and obtain N 1A lithium-ion battery state of health estimation model. Taking the local voltage capacity increment as the feature of the lithium-ion battery state of health estimation model has the advantages of being easy to calculate and not requiring complex filtering algorithm processing. In addition, each lithium-ion battery state of health estimation model corresponds to its own voltage range, which enables the lithium-ion battery state of health estimation model to determine which several lithium-ion battery state of health estimation models can be put into use and perform joint estimation according to the actual voltage sampling data in application, greatly improving the flexibility of lithium-ion battery state of health estimation.

[0046] The root mean square error loss function δ in the present invention a is:

[0047]

[0048] where M val is the number of charging data samples in the a-th voltage range, u is the index of M val , y val (u) represents the true value of the battery state of health of the u-th charging data sample, represents the estimated value of the battery state of health of the a-th voltage range with respect to the u-th charging data sample.

[0049] S5. Real-time collect the lithium battery charging segment at the current charge and discharge cycle number, select the corresponding lithium-ion battery state of health estimation model according to the charging segment, obtain the multi-model joint estimation value, and obtain the optimal estimation of the current lithium battery state of health through the Kalman filtering algorithm. Considering that in the actual application of multi-model joint estimation, in extreme cases, when the charging voltage range is too small in a certain cycle and only individual lithium-ion battery state of health estimation models can be called for estimation, and the accuracy of this lithium-ion battery state of health estimation model is poor, it will lead to a large deviation in the estimation result of this round. By introducing the Kalman filtering algorithm, the estimation value of the previous round can be combined with the joint estimation result of this round of models, and the estimation error of this round can be adjusted to give a trade-off value between the two, reducing the deviation brought by extreme cases, thereby improving the estimation accuracy of the lithium-ion battery state of health estimation. Specifically, it includes the following sub-steps:

[0050] S5.1. In practical applications, only a small range of voltage data can often be collected, so only a few voltage ranges can obtain the required features and give estimation results. Therefore, by real-time collecting the lithium battery charging segment at the current charge and discharge cycle number, selecting the corresponding lithium-ion battery state of health estimation model according to the charging segment, calculating the root mean square error of each lithium-ion battery state of health estimation model, and setting the weight w of each lithium-ion battery state of health estimation model according to the root mean square error j :

[0051]

[0052] Among them, J is the number of lithium-ion battery state-of-health estimation models corresponding to the charging segments, j represents the index of J, and δ j is the root mean square error between the j-th lithium-ion battery state-of-health estimation model and the charging segment of the real-time lithium battery;

[0053] S5.2. Estimate the state of health of the lithium battery through the corresponding lithium-ion battery state-of-health estimation model, and combine the weights of each lithium-ion battery state-of-health estimation model to obtain a multi-model joint estimation value

[0054]

[0055] Among them, is a matrix set composed of the state of health of the lithium battery estimated by each lithium-ion battery state-of-health estimation model, and W is the weight set of the lithium-ion battery state-of-health estimation model;

[0056] S5.3. Further correct the multi-model joint estimation value using the Kalman filter to obtain the optimal estimation of the current state of health of the lithium battery

[0057] Among them, is the prior estimate of the current charge-discharge cycle number k, and its initial value is K(k) is the Kalman gain of the current charge-discharge cycle number k, P - (k) is the prior estimate variance of the current charge-discharge cycle number k, and P - (k) = P(k - 1) + Q; P(k - 1) is the optimal estimate variance at the k - 1 charge-discharge cycle number, and its initial value is P(1) = R(1); Q is the process variance, R(k) is the estimate variance of the current charge-discharge cycle number k, n(k) is the number of lithium-ion battery state-of-health estimation models used at the current charge-discharge cycle number k, and J(k) is the lithium-ion battery state-of-health estimation model used at the current charge-discharge cycle number k.

[0058] The lithium battery state of health estimation method based on local capacity increment features of the present invention is based on a support vector regression model, takes the local voltage capacity interval as its input, creates multiple lithium-ion battery state of health estimation models for joint estimation, and greatly improves the flexibility of the application of the lithium-ion battery state of health estimation model. In addition, the Kalman filter algorithm is introduced in the present invention, and by reasonably setting its process variance and dynamically adjusting its estimation error, the accuracy of the multi-model joint estimation is improved. When the lithium battery state of health estimation method of the present invention is used for testing the Oxford University battery aging public dataset, with the first lithium battery as the training set and the remaining 7 lithium batteries as the test set, the root mean square errors when using the battery capacity as the estimation result are respectively: 1.17%, 1.16%, 1.02%, 1.44%, 1.21%, 1.44%, 1.44%, while the error of the usual lithium-ion battery state of health estimation is within 3%-5%, indicating that the lithium battery state of health estimation method based on local capacity increment features of the present invention has high accuracy.

[0059] The above is only the preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiment. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, several improvements and refinements made without departing from the principle of the present invention should be regarded as within the protection scope of the present invention.

Claims

1. A method for estimating the state of health of a lithium battery based on local capacity increment characteristics, characterized in that, it specifically includes the following steps: S1. Obtain the charging data of each charge-discharge cycle during the process of the lithium battery from factory to retirement. Taking the first charge in the charge-discharge cycle as the standard, evaluate the voltage range corresponding to the state of charge from 20% to 80%, and retain all the charging data within the voltage range in the charging data; S2. Divide the voltage range in the charging data retained in each round by an interval ΔU 1 into N 1 voltage intervals, and perform voltage repair on the N 1 voltage intervals; S3. Divide each voltage range for voltage restoration into N voltage sub-ranges at intervals of ΔU, and calculate the corresponding local voltage capacity increment for each sub-range; 2 Divide into N 2 voltage sub-ranges, and calculate the corresponding local voltage capacity increment for each sub-range; S4. Input the local voltage capacity increment corresponding to each sub-interval into the support vector regression model for training until the root mean square error loss function converges, and obtain N 1 lithium-ion battery state of health estimation models; S5. Real-time collect the charging segment of the lithium battery at the current number of charge-discharge cycles, select the corresponding lithium-ion battery state of health estimation model according to the charging segment, obtain the multi-model joint estimation value, and obtain the optimal estimation of the current lithium battery state of health through the Kalman filter algorithm; including the following sub-steps: S5.

1. Real-time collect the lithium battery charging segments at the current charge-discharge cycle number, select the corresponding lithium-ion battery health state estimation model according to the charging segments, calculate the root mean square error of each lithium-ion battery health state estimation model, and set the weight w of each lithium-ion battery health state estimation model according to the root mean square error j : Among them, J is the number of lithium-ion battery state of health estimation models corresponding to the charging segment, j represents the index of J, and δ j is the root mean square error between the j-th lithium-ion battery state of health estimation model and the charging segment of the real-time lithium battery; S5.

2. Estimate the state of health of the lithium battery through the corresponding state-of-health estimation model of the lithium-ion battery, and combine the weights of each state-of-health estimation model of the lithium-ion battery to obtain a multi-model joint estimation value Among them, is a matrix set composed of the health states of lithium batteries estimated by each lithium-ion battery health state estimation model, and W is a weight set of the lithium-ion battery health state estimation models; S5.

3. Further correct the joint estimation value of multiple models using the Kalman filter to obtain the optimal estimation of the current state of health of the lithium battery Among them, is the prior estimate of the current charge-discharge cycle number k, and K(k) is the Kalman gain of the current charge-discharge cycle number k. P - (k) is the prior estimate variance of the current charge-discharge cycle number k, and P - (k) = P(k - 1) + Q, P(k - 1) is the optimal estimate variance at the k - 1 charge-discharge cycle number, Q is the process variance, and R(k) is the estimate variance of the current charge-discharge cycle number k.

2. The method for estimating the state of health of a lithium battery based on local capacity increment characteristics according to claim 1, characterized in that, The interval ΔU 1 is 50 mV - 200 mV.

3. The method for estimating the state of health of a lithium battery based on local capacity increment characteristics according to claim 1, characterized in that, The interval ΔU 2 is 3 mV - 10 mV.

4. The method for estimating the state of health of a lithium battery based on local capacity increment characteristics according to claim 1, characterized in that, For N 1 The process of voltage restoration for a voltage range is as follows: The charging voltage curve increases monotonically with the state of charge. If the voltage value at any moment during charging is lower than all previous moments, then the voltage value at that moment is abnormal. Find the normal voltages at the two adjacent moments on the left and right of the abnormal voltage value at that moment, and replace the abnormal voltage value by linear interpolation: Among them, t abnormal is the moment when the abnormal voltage appears, is the moment immediately adjacent to the left of the abnormal voltage, is the moment immediately adjacent to the right of the abnormal voltage, is the corresponding normal voltage value, is the corresponding normal voltage value, V renew is the repair value for the abnormal voltage.

5. The method for estimating the state of health of a lithium battery based on local capacity increment characteristics according to claim 1, characterized in that, the local capacity increment Δq corresponding to the sub-interval is: where \(i(t)\) is the current corresponding to the lithium battery at time \(t\) during the charging process, \(t\) st is the starting time of the sub - interval, and \(t\) ed is the ending time of the sub - interval.

6. The method for estimating the state of health of a lithium battery based on local capacity increment characteristics according to claim 1, characterized in that, The root mean square error loss function δ a is as follows: Among them, M val is the number of charging data samples in the a-th voltage interval, u is the index of M val , y val (u) represents the true value of the state of health of the battery for the u-th charging data sample, represents the estimated value of the state of health of the battery for the u-th charging data sample in the a-th voltage interval.

Images