A method for estimating the state of charge (SOH) of lithium-ion batteries based on random short-term charging data

Abstract

This invention relates to a lithium-ion battery SOH estimation method based on random short-term charging data, comprising the following steps: Step 1, data acquisition; Step 2, polynomial fitting; Step 3, short-term charging voltage estimation; Step 4, obtaining optimal coefficients; Step 5, complete charging voltage estimation; Step 6, data processing; Step 7, basic model construction; Step 8, battery SOH estimation: using the equal-voltage charging time as model input, the estimated battery SOH is obtained. Based on the single hidden layer structure and fast learning characteristics of Extreme Learning Machine (ELM), the overfitting problem easily caused by complex algorithms is avoided. In step 7, model training only used 40% of the data to achieve good prediction results, proving that this method has strong learning ability and does not require a large amount of training data.

Description

Technical Field

[0001] This invention belongs to the field of lithium-ion battery technology, and particularly relates to a method for estimating the state of charge (SOH) of lithium-ion batteries based on random short-term charging data. Background Technology

[0002] Lithium-ion batteries are widely used in energy storage systems due to their advantages such as high energy density, small size, no memory effect, low self-discharge rate, and environmental friendliness. Accurate prediction of battery state of health (SOH) can greatly contribute to battery health management, preventing safety accidents caused by overuse and extending battery lifespan. SOH is also an important reference for SOC estimation.

[0003] However, lithium-ion batteries are complex physicochemical reaction systems. Existing machine learning methods often require large amounts of training data. Short-term random charging data cannot provide sufficient features, feature acquisition is complex, model adaptability is poor, and repeated model training is necessary. Even well-trained models often have poor prediction accuracy. Therefore, how to obtain highly adaptable, computationally efficient, and effective features based on short-term random charging data to achieve accurate estimation of the state of health (SOH) of lithium-ion batteries is a current challenge in lithium SOH estimation. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a lithium-ion battery SOH estimation method based on random short-term charging data to solve the problems of existing technologies, such as the need for a large amount of training data, the inability to obtain features from short-term random charging data, the complexity of feature acquisition, poor model adaptability, the need for repeated model training, and poor prediction accuracy of the trained model.

[0005] To solve the above technical problems, the technical solution of the present invention is as follows: A method for estimating the state of charge (SOH) of a lithium-ion battery based on random short-term charging data is provided. Its innovation lies in the following steps:

[0006] Step 1, Data Acquisition: Perform cycle charge-discharge tests on the lithium-ion battery until the discharge capacity is lower than 80% of the nominal capacity, and record the battery charging voltage and discharge capacity data in real time.

[0007] Step 2, Polynomial Fitting: Perform voltage polynomial fitting on the charging voltage data obtained in Step 1 during the constant current charging stage with equal cycle intervals to obtain n constant current charging voltage fitting polynomial curves V1-V n , where the value of n ranges from 5 to 15;

[0008] Step 3, Short-time charging voltage estimation: Obtain the real-time constant current charging voltage data V0 at 15% SOC length, and select the constant current charging voltage obtained in Step 2 to fit the polynomial curve V1-V. n ΔV1-ΔV at the same time pointm Through coefficient combination ΔV1-ΔV m A rough estimate of the voltage V for this segment can be obtained. e ;

[0009] Step 4: Obtain the optimal coefficients: Using the whale optimization algorithm, obtain ΔV1-ΔV from Step 3. m The optimal coefficient a1-a m ;

[0010] Step 5, Complete charging voltage estimation: Using the obtained optimal coefficient a1-a m The corresponding V1-V combination m This yields the complete voltage estimation curve V;

[0011] Step 6, Data Processing: Preprocess the charging voltage data obtained in Step 1 or the voltage estimation data and discharge capacity data obtained in Step 5 to obtain the isovoltage charging time and SOH data for a specific voltage range.

[0012] Step 7, Basic Model Construction: Take the first 40% of the SOH obtained in Step 6, and the isovoltage charging time data, run the extreme learning machine method to find the correlation, and obtain the trained basic model;

[0013] Step 8: Battery SOH estimation: Use the equal voltage charging time as the model input to obtain the estimated battery SOH.

[0014] Furthermore, the voltage polynomial fitting method with equal cyclic intervals in step 2 is as follows:

[0015] Read the lithium-ion battery charging voltage and divide it into n-1 intervals according to the number of cycles, with n corresponding boundary cycles;

[0016] The constant current charging voltage of n boundary cycles within n-1 intervals is fitted using a polynomial. The specific fitting formula is as follows:

[0017]

[0018] In the formula, V n To fit the constant current charging voltage curves for each boundary cycle, SOC represents the corresponding state of charge, and α i These are the polynomial coefficients.

[0019] Furthermore, in step 3, the real-time voltage data V0 of constant current charging at 15% SOC length is obtained, and the constant current charging voltage fitting polynomial curve V1-V obtained in step 2 is selected. n ΔV1-ΔV at the same time point m Through coefficient combination ΔV1-ΔV m A rough estimate of the voltage V for this segment can be obtained.e The specific method is as follows:

[0020] Select V1-V n In the process, the ΔV1-ΔV corresponds to the time period of the real-time acquired constant current charging 15% SOC data. m Where m≤n, the corresponding time period of this data segment may not have the same SOC. When the charging time is different due to aging, the polynomial that does not meet the conditions is discarded.

[0021] Using the m data segments obtained through coefficient combinations, a new data segment is obtained, which is the inaccurate estimated voltage for that segment. The formula is as follows:

[0022] V e = a1ΔV1 + a2ΔV2 + ... + a m ΔV m

[0023] In the formula, V e To estimate the voltage for the 15% SOC charging stage, a m is a coefficient.

[0024] Furthermore, in step 4, the whale optimization algorithm is used to obtain ΔV1-ΔV from step 3. m The optimal coefficient a1-a m The specific method is as follows:

[0025] Calculate V e The root mean square error (RMSE) between the true value and the actual value, and V e The absolute value K of the difference between the slopes of the lines connecting the beginning and end of the two curve segments relative to the true value;

[0026] Taking RMSE and K, the objective function of the whale optimization algorithm is constructed as follows:

[0027] obj = RMSE / (1-K)

[0028] Using the whale optimization algorithm, with boundary conditions set to [0,1], the optimal solution of the objective function is obtained, yielding ΔV1-ΔV. m The optimal coefficient a1-a m .

[0029] Furthermore, in step 5, the obtained optimal coefficient a1-a is used. m The corresponding V1-V combination m The estimation formula for obtaining the complete voltage estimation curve V is:

[0030] V = a1V1 + a2V2 + ... + a m V m

[0031] In the formula, V represents the estimated constant current charging voltage curve, and a m These are the optimal coefficients obtained in step 4.

[0032] Furthermore, the preprocessing method in step 6 is as follows:

[0033] Read the charging voltage and discharging capacity of the lithium-ion battery;

[0034] The charging time within a fixed charging voltage range is taken as the charging time for a specific voltage range.

[0035] The current battery SOH is obtained by dividing the discharge capacity by the nominal battery capacity.

[0036] Furthermore, in step 7, the Extreme Learning Machine (ELM) method is used to find associations, and its ELM output can be expressed as:

[0037]

[0038] In the formula ω i Let b be the weight vector connecting the input neuron and the i-th hidden neuron. i Let g be the bias of the i-th hidden neuron, g be the activation function, and β be the bias. i Let N be the weight vector of the i-th hidden neuron and the output neuron. neuron This represents the number of neurons.

[0039] Compared with existing technologies, the lithium-ion battery SOH estimation method based on random short-term charging data of the present invention has the following advantages:

[0040] (1) Based on the single hidden layer structure and fast learning characteristics of the Extreme Learning Machine, the overfitting problem that is easily caused by complex algorithms is avoided. In step 7, only 40% of the data was used for training, and good prediction results were achieved, proving that the method has strong learning ability and does not require a large amount of training data.

[0041] (2) Steps 2-5 realize the estimation of charging curve using random short-term charging data, and then extract the equal voltage charging time as a feature input to the SOH prediction model to complete capacity prediction, thus solving the research difficulty that cannot obtain features in short-term random charging.

[0042] (3) In the feature extraction step in step 6, only the equal voltage charging time is used, avoiding the filtering and other processing of the acquired data, avoiding a large amount of calculation in the feature extraction step, and realizing online estimation of SOH.

[0043] (4) In step 7, the basic model is constructed using the Extreme Learning Machine (ELM) method, which has strong learning and generalization capabilities. The constructed SOH prediction model is highly adaptable, has high prediction accuracy, and does not require repeated training. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the implementation process of the present invention.

[0046] Figure 2 This is the estimated charging voltage obtained from the implementation of this invention.

[0047] Figure 3 This is the estimated SOH result of the lithium-ion battery obtained from the implementation of this invention. Detailed Implementation

[0048] To enable those skilled in the art to better understand the present invention, the present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0049] This invention provides a method for estimating the state of charge (SOH) of a lithium-ion battery based on random short-term charging data, such as... Figure 1 As shown, it includes the following steps:

[0050] Step 1, Data Acquisition: Perform cycle charge-discharge tests on the lithium-ion battery until the discharge capacity is lower than 80% of the nominal capacity, and record the battery charging voltage and discharge capacity data in real time.

[0051] The test process involved fully charging the battery using constant current and constant voltage, letting it stand for 1 hour, then discharging it to the battery cutoff voltage, and letting it stand for another hour. This process was repeated, and the experimental current, voltage, and discharge capacity were recorded. The experimental data are shown in Table 1.

[0052] Table 1

[0053] Current (A) Voltage (V) Capacity(Ah) SOH 2.000200 3.4898 0.000000 / 2.001400 3.5218 0.016678 / 2.001400 3.5301 0.033357 / …… …… …… / 2.001400 3.5258 0.116753 / 2.001400 3.5295 0.133432 / 2.001400 3.5329 0.150111 / 2.002700 3.5367 0.166790 / …… …… …… / -4.002900 3.2840 3.569210 / -4.002900 3.2598 3.602568 / -4.002900 3.2204 3.635925 / -4.002900 3.1324 3.669281 0.9173

[0054] Step 2, Polynomial Fitting: Perform voltage polynomial fitting on the charging voltage data obtained in Step 1 during the constant current charging stage with equal cycle intervals to obtain n constant current charging voltage fitting polynomial curves V1-V n Where n ranges from 5 to 15; the voltage polynomial fitting method with equal cyclic intervals is as follows:

[0055] Read the lithium-ion battery charging voltage and divide it into n-1 intervals according to the number of cycles, with n corresponding boundary cycles;

[0056] The constant current charging voltage of n boundary cycles within n-1 intervals is fitted using a polynomial. The specific fitting formula is as follows:

[0057]

[0058] In the formula, V n To fit the constant current charging voltage curves for each boundary cycle, SOC represents the corresponding state of charge, and α i These are the polynomial coefficients.

[0059] Step 3, Short-time charging voltage estimation: Obtain the real-time constant current charging voltage data V0 at 15% SOC length, and select the constant current charging voltage obtained in Step 2 to fit the polynomial curve V1-V. n ΔV1-ΔV at the same time point m Through coefficient combination ΔV1-ΔV m A rough estimate of the voltage V for this segment can be obtained. e The specific method is as follows:

[0060] Select V1-V n In the process, the ΔV1-ΔV corresponds to the time period of the real-time acquired constant current charging 15% SOC data. m Where m≤n, the corresponding time period of this data segment may not have the same SOC. When the charging time is different due to aging, the polynomial that does not meet the conditions is discarded.

[0061] Using the m data segments obtained through coefficient combinations, a new data segment is obtained, which is the inaccurate estimated voltage for that segment. The formula is as follows:

[0062] V e = a1ΔV1 + a2ΔV2 + ... + a m ΔV m

[0063] In the formula, V e To estimate the voltage for the 15% SOC charging stage, a m is a coefficient.

[0064] Step 4: Obtain the optimal coefficients: Using the whale optimization algorithm, obtain ΔV1-ΔV from Step 3. m The optimal coefficient a1-a m ;

[0065] Using the whale optimization algorithm, obtain ΔV1-ΔV in step 3. m The optimal coefficient a1-a m The specific method is as follows:

[0066] Calculate V e The root mean square error (RMSE) between the true value and the actual value, and V e The absolute value K of the difference between the slopes of the lines connecting the beginning and end of the two curve segments relative to the true value;

[0067] Taking RMSE and K, the objective function of the whale optimization algorithm is constructed as follows:

[0068] obj = RMSE / (1-K)

[0069] Using the whale optimization algorithm, with boundary conditions set to [0,1], the optimal solution of the objective function is obtained, yielding ΔV1-ΔV. m The optimal coefficient a1-a m .

[0070] Step 5, Complete charging voltage estimation: Using the obtained optimal coefficient a1-a m The corresponding V1-V combination m The complete voltage estimation curve V is obtained; the estimation formula for the complete voltage estimation curve V is:

[0071] V = a1V1 + a2V2 + ... + a m V m

[0072] In the formula, V represents the estimated constant current charging voltage curve, and a m These are the optimal coefficients obtained in step 4.

[0073] The estimation results are as follows Figure 2 As shown, the estimated voltage matches the actual voltage very well, proving that the voltage estimation method has good voltage estimation capability.

[0074] Step 6, Data Processing: Preprocess the charging voltage data obtained in Step 1 or the voltage estimation data and discharge capacity data obtained in Step 5 to obtain the isovoltage charging time and SOH data for a specific voltage range.

[0075] During implementation, the charging voltage data obtained in step 1 or the estimated charging voltage data obtained in step S5 are used to determine the charging time for the three voltage ranges [3.6V, 3.7V], [3.7V, 3.8V], and [3.6V, 3.8V] as the isovoltage charging time T for a specific voltage range. i = [t1, t2, t3], where i represents the cycle number;

[0076] Normalize the charging data under different cycle lives, extract the discharge capacity, and calculate the corresponding SOH using the following formula to obtain the SOH data:

[0077]

[0078] In the formula, Ca is the current discharge capacity, Ca N This refers to the nominal capacity of a lithium-ion battery.

[0079] The obtained charging time T i The dataset was created by combining the SOH data, and the experimental data are shown in Table 2:

[0080] Table 2

[0081] Loop count Charging time SOH Training set / test 1 <![CDATA[T1]]> 0.9229 training set 2 <![CDATA[T2]]> 0.9271 training set 3 <![CDATA[T3]]> 0.9275 training set …… …… …… training set 360 <![CDATA[T 360 ]]> 0.8445 training set 361 <![CDATA[T 361 ]]> 0.8443 test set …… …… …… test set 901 <![CDATA[T 901 ]]> 0.6941 test set 902 <![CDATA[T 902 ]]> 0.6990 test set 903 <![CDATA[T 903 ]]> 0.6944 test set 904 <![CDATA[T 904 ]]> 0.6935 test set

[0082] Step 7, Basic Model Construction: Take the first 40% of the SOH obtained in Step 6, and the isovoltage charging time data, run the extreme learning machine method to find the correlation, and obtain the trained basic model;

[0083] Initialize the parameters of the extreme learning machine method;

[0084] The training set obtained in step 6 is input into the Extreme Learning Machine (ELM) model. The SOH value of the ELM is calculated by the ELM, and the output of the ELM can be expressed as:

[0085]

[0086] In the formula ω i Let b be the weight vector connecting the input neuron and the i-th hidden neuron. i Let g be the bias of the i-th hidden neuron, g be the activation function, and β be the bias. i Let N be the weight vector of the i-th hidden neuron and the output neuron. neuron This represents the number of neurons.

[0087] Step 8: Using the charging time of the test set in Table 2 as model input, obtain the estimated battery SOH. The estimation results are as follows: Figure 3 As shown, the estimated value is closely distributed around the true value, indicating that the established SOH estimation model can perform the SOH estimation work of lithium battery very well.

Claims

1. A method for estimating the state of charge (SOH) of a lithium-ion battery based on random short-term charging data, characterized in that, Includes the following steps: Step 1, Data Acquisition: Perform cycle charge-discharge tests on the lithium-ion battery until the discharge capacity is lower than 80% of the nominal capacity, and record the battery charging voltage and discharge capacity data in real time. Step 2, Polynomial Fitting: Perform voltage polynomial fitting on the charging voltage data obtained in Step 1 during the constant current charging stage with equal cycle intervals to obtain n constant current charging voltage fitting polynomial curves V1-V n , where the value of n ranges from 5 to 15; Step 3, Short-time charging voltage estimation: Obtain the real-time constant current charging voltage data V0 at 15% SOC length, and select the constant current charging voltage obtained in Step 2 to fit the polynomial curve V1-V. n V1-ΔV at the same time point m Through coefficient combination ΔV1-ΔV m A rough estimate of the voltage V for this segment is obtained. e The specific method is as follows: Select V1-V n In the process, the ΔV1-ΔV corresponds to the time period of the real-time acquired constant current charging 15% SOC data. m Where m≤n, the corresponding time period of this data segment may not have the same SOC. When the charging time is different due to aging, the polynomial that does not meet the conditions is discarded. Using the m data segments obtained through coefficient combinations, a new data segment is obtained, which is the inaccurate estimated voltage for that segment. The formula is as follows: V e =a1ΔV1+a2ΔV2+…+a m ΔV m In the formula, V e To estimate the voltage for the 15% SOC charging stage, a m For coefficients; Step 4: Obtain the optimal coefficients: Using the whale optimization algorithm, obtain ΔV1-ΔV from Step 3. m The optimal coefficient a1-a m ; Step 5, Complete charging voltage estimation: Using the obtained optimal coefficient a1-a m The corresponding V1-V combination m This yields the complete voltage estimation curve V; Step 6, Data Processing: Preprocess the charging voltage data obtained in Step 1 or the voltage estimation data and discharge capacity data obtained in Step 5 to obtain the isovoltage charging time and SOH data for a specific voltage range. Step 7, Basic Model Construction: Take the first 40% of the SOH obtained in Step 6, and the isovoltage charging time data, run the extreme learning machine method to find the correlation, and obtain the trained basic model; Step 8: Battery SOH estimation: Use the equal voltage charging time as the model input to obtain the estimated battery SOH.

2. The lithium-ion battery SOH estimation method based on short-term random charging data according to claim 1, characterized in that, The voltage polynomial fitting method with equal cyclic intervals in step 2 is as follows: Read the lithium-ion battery charging voltage and divide it into n-1 intervals according to the number of cycles, with n corresponding boundary cycles; The constant current charging voltage of n boundary cycles within n-1 intervals is fitted using a polynomial. The specific fitting formula is as follows: In the formula, V n To fit the constant current charging voltage curves for each boundary cycle, SOC represents the corresponding state of charge, and α i These are the polynomial coefficients.

3. The lithium-ion battery SOH estimation method based on short-term random charging data according to claim 1, characterized in that, In step 3, the real-time voltage data V0 of constant current charging at 15% SOC length is obtained, and the constant current charging voltage fitting polynomial curve V1-V obtained in step 2 is selected. n ΔV1-ΔV at the same time point m Through the coefficient combination ΔV1-ΔV m A rough estimate of the voltage V for this segment is obtained. e The specific method is as follows: Select V1-V n In the process, the ΔV1-ΔV corresponds to the time period of the real-time acquired constant current charging 15% SOC data. m Where m≤n, the corresponding time period of this data segment may not have the same SOC. When the charging time is different due to aging, the polynomial that does not meet the conditions is discarded. Using the m data segments obtained through coefficient combinations, a new data segment is obtained, which is the inaccurate estimated voltage for that segment. The formula is as follows: V e =a1ΔV1+a2ΔV2+…+a m ΔV m In the formula, V e To estimate the voltage for the 15% SOC charging stage, a m is a coefficient.

4. The lithium-ion battery SOH estimation method based on short-term random charging data according to claim 1, characterized in that, In step 4, the whale optimization algorithm is used to obtain V1-ΔV from step 3. m The optimal coefficient a1-a m The specific method is as follows: Calculate V e The root mean square error (RMSE) between the true value and the actual value, and V e The absolute value K of the difference between the slopes of the lines connecting the beginning and end of the two curve segments relative to the true value; Taking RMSE and K, the objective function of the whale optimization algorithm is constructed as follows: obj = RMSE / (1-K) Using the whale optimization algorithm, with boundary conditions set to [0,1], the optimal solution of the objective function is obtained, yielding V1-ΔV. m The optimal coefficient a1-a m .

5. The lithium-ion battery SOH estimation method based on short-term random charging data according to claim 1, characterized in that, In step 5, the obtained optimal coefficient a1-a is used m The corresponding V1-V combination m The estimation formula for obtaining the complete voltage estimation curve V is: V=a1V1+a2V2+…+a m V m In the formula, V represents the estimated constant current charging voltage curve, and a m These are the optimal coefficients obtained in step 4.

6. The lithium-ion battery SOH estimation method based on short-term random charging data according to claim 1, characterized in that, The preprocessing method in step 6 is as follows: Read the charging voltage and discharging capacity of the lithium-ion battery; The charging time within a fixed charging voltage range is taken as the charging time for a specific voltage range. The current battery SOH is obtained by dividing the discharge capacity by the nominal battery capacity.

7. The lithium-ion battery SOH estimation method based on short-term random charging data according to claim 1, characterized in that, In step 7, the Extreme Learning Machine (ELM) method is used to find associations. The output of the ELM can be expressed as: In the formula ω i Let b be the weight vector connecting the input neuron and the i-th hidden neuron. i Let g be the bias of the i-th hidden neuron, g be the activation function, and β be the bias. i Let N be the weight vector of the i-th hidden neuron and the output neuron. neuron This represents the number of neurons.

Images