Lithium ion battery residual capacity prediction method based on partial charging curve

Abstract

The current prediction method ignores the charging habit of a user, so that feature extraction is difficult, and capacity estimation is inaccurate. The invention provides a lithium ion battery residual capacity prediction method based on a partial charging curve. The method comprises the following steps: firstly, analyzing coverage probabilities of various SOC charging processes based on real driving data, and extracting charging duration under equal voltage as health degree features; then, according to the Pearson correlation coefficient, the correlation between the health features and the residual capacity is calculated; and then, in combination with the cumulative probability density function and the correlation, establishing a joint optimization objective function, and determining an optimal SOC range by using a particle swarm optimization algorithm. And finally, providing a time convolutional network model to capture an implicit nonlinear relationship between the charging time and the residual capacity in the optimal SOC range. Cross validation shows that the method can accurately track the reference capacity, and the estimation error is within 2.3% of the nominal capacity.

Description

Technical Field

[0001] The invention relates to the technical field of lithium ion batteries, and in particular to a method for predicting the remaining capacity of a lithium ion battery based on a partial charging curve. Background Art

[0002] As a key component of energy storage and electric vehicle power source, battery performance and cycle life have attracted much attention. As one of the key parameters inside lithium-ion batteries, accurate estimation of battery remaining capacity helps avoid excessive use of batteries and ensure battery safety. At the same time, it can also provide a reference for battery state of charge estimation.

[0003] However, due to the complexity of the chemical reactions inside the battery, it is difficult for existing methods to guarantee their estimation accuracy. At present, the estimation of the remaining capacity of lithium-ion batteries is mainly based on machine learning methods, but this method needs to first extract appropriate health features and then use advanced machine learning methods to estimate the remaining capacity. The existing health feature extraction method usually requires complete charging or discharging process data, which cannot be met in actual use; since the user's random charging behavior may not be within the specified SOC range, this may make it impossible to obtain health features and thus impossible to estimate capacity. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method for predicting the remaining capacity of a lithium-ion battery based on a partial charging curve, which has the advantages of high model fitting degree and high estimation accuracy, and solves the above-mentioned technical problems.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for predicting the remaining capacity of a lithium-ion battery based on a partial charging curve, comprising the following steps:

[0006] S1. Collect data during the actual vehicle cycle charging process;

[0007] The data of the actual vehicle cycle charging process is collected from 30 electric bicycles randomly selected from the big data platform at a sampling interval of 10 seconds, including the battery charging voltage, charging time, charging start state of charge and charging end state of charge data;

[0008] S2. Analyze and count the collected charging data. The specific steps are as follows:

[0009] S2.1. Extract the frequency distribution of the initial and final charging state of charge from the collected data, establish a clustered bar chart with the battery state of charge as the horizontal axis and the number of occurrences as the vertical axis, and obtain the cumulative probability distribution curve of starting or stopping charging SOC;

[0010] S2.2. In order to further characterize the probability of different SOCs during the charging process, the probability distribution of the initial and final charging SOCs is calculated using the cumulative probability distribution. The specific expression is as follows:

[0011]

[0012] Where f(SOC) represents the cumulative probability distribution function of starting or stopping charging SOC, and v represents the battery voltage;

[0013] According to the calculation results, the cumulative probability distribution curve of the voltage corresponding to the start-stop charging state of charge is drawn as follows: Figure 3 As shown;

[0014] S3. Extract the data of the battery charging process as health features during the cycle. The specific steps are as follows:

[0015] S3.1. The charging duration within the SOC range is used as a health feature to characterize the battery capacity. In order to minimize the capacity estimation error, the Pearson correlation coefficient is used to quantify the relationship between the health feature and the capacity. The specific expression is as follows:

[0016]

[0017] In the formula, x i represents the i-th feature, and They represent the average values ​​of characteristic sequence and capacity sequence respectively, and z represents capacity;

[0018] S3.2. Taking the cumulative probability density of the SOC range as the optimization target, the joint optimization objective function is obtained. The specific expression is as follows:

[0019]

[0020] In the formula, α and β are and the weight of the cumulative probability distribution, and β = 1-α, α is set to 0.5;

[0021] S3.3, in order to solve the objective function, the particle swarm optimization algorithm is used to minimize the objective function. The specific expression is as follows:

[0022]

[0023] Where i = 1, 2, ..., N, N is the total number of particles in the group, represents the position of particle i at the kth iteration, represents the velocity of particle i at the kth iteration, represents the speed of particle i at the k+1th iteration, ω is called the inertia factor, and its value is a non-negative number, c1 represents the individual optimal value learning factor, c2 represents the group optimal value learning factor, c1=c2=2, rand1() and rand2() represent random numbers uniformly distributed between [0, 1], rand1() may be equal to rand2() or not, represents the individual optimal position of particle i at the kth iteration, gbest k It represents the global optimal position of the entire particle swarm up to the kth iteration, and finally obtains the optimal SOC range, which is [4.083, 4.185];

[0024] S4. Establish a remaining capacity estimation model based on health characteristics;

[0025] S4.1. Establish TCN model;

[0026] The temporal convolutional network (TCN) model is established, and the TCN model consists of causal convolution, dilated convolution and residual block. Causal convolution is used to prevent information leakage. The residual block has two dilated causal convolution layers, two ReLU layers, two normalization layers, and two Dropout layers. The one-dimensional full convolution network is implemented by padding the length to zero (kernel size is -1). The causal convolution operation at X is defined as:

[0027]

[0028] Where T is the size of the convolution kernel, t=1 means that the value of t starts from 1, t∈[1,T], x t is the value of the input time series at time t;

[0029] In addition to causal convolution, TCN also uses dilated convolution. t The dilated convolution operation at is defined as:

[0030]

[0031] Where X=(x1,x2,...,x t-1 ,x T ) is a one-dimensional sequence input, f=f0,f1,…,f k-1 Indicates filter, * d represents a dilated convolution operation with a dilation factor d, k is the filter size, k-1-id represents the past direction, and the receptive field is increased by increasing the dilation factor d. The value of the dilation factor is set to 2i;

[0032] S4.2, taking the charging time corresponding to the optimal SOC range as the input of the TCN neural network, and the corresponding remaining battery capacity as the output of the TCN network to complete the model construction;

[0033] The optimal SOC range is [4.083, 4.185];

[0034] S5. Estimate and verify the remaining capacity of the battery through the established model;

[0035] S5.1. Use the established model to estimate the remaining capacity of the battery and perform cross-validation;

[0036] In order to verify the proposed health feature optimization method and capacity estimation results, cross validation was performed using the Oxford University battery dataset;

[0037] S5.2. Analyze the root mean square error (RMSE), maximum absolute error (MAE) and average absolute error (AAE) of each battery. The specific calculation formula is as follows:

[0038]

[0039]

[0040]

[0041] In the formula, y i represents the true value, Represents the predicted value, m represents the number of instances in the data set, and MAX() represents the maximum value.

[0042] Beneficial effects of the present invention

[0043] The present invention reads the user's charging data, counts the number of occurrences of the starting state of charge and the stopping state of charge, and uses CPD to draw a cumulative probability distribution curve, so that the user's random charging behavior can extract accurate health characteristics. Then, combined with the cumulative probability density function and correlation, a joint optimization objective function is established, and the particle swarm optimization algorithm is used to determine the optimal SOC range. The battery health value is predicted according to the established TCN model and cross-validated using the Oxford University battery data set, thereby completing the accurate estimation of the remaining capacity of the lithium battery. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the process of the present invention;

[0045] Figure 2 To start charging the state of charge distribution;

[0046] Figure 3To stop charging the state of charge distribution;

[0047] Figure 4 CPD curve for starting and stopping charging voltage;

[0048] Figure 5 is the capacity prediction result and error of battery 1;

[0049] Figure 6 The capacity prediction result and error of battery 2. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] See also Figure 1 , a method for predicting the remaining capacity of a lithium-ion battery based on a partial charging curve, comprising the following steps:

[0052] S1. Collect data during the actual vehicle cycle charging process;

[0053] The data of the actual vehicle cycle charging process is collected from 30 electric bicycles randomly selected from the big data platform at a sampling interval of 10 seconds, including the battery charging voltage, charging time, charging start state of charge and charging end state of charge data;

[0054] In this embodiment, the monitored battery pack consists of 14 modules, each module consists of 6 batteries connected in parallel, and 3 thermistors are evenly attached inside the battery pack. A three-stage CC charging strategy is used to charge the battery pack of the electric scooter. First, the battery pack is charged at 0.1C until the voltage of all batteries exceeds 3.6V, and then a 0.2C current is applied to charge the battery until the voltage of any battery reaches 4.19V. Finally, a 0.05C current is charged until the voltage of any battery reaches 4.2V;

[0055] S2. Analyze and count the collected charging data. The specific steps are as follows:

[0056] S2.1. Extract the frequency distribution of the initial and final charging state of charge in the collected data, establish a clustered bar chart with the battery state of charge as the horizontal axis and the number of occurrences as the vertical axis, and obtain the cumulative probability distribution curve of starting or stopping charging SOC, such as Figure 2 and Figure 3 As shown;

[0057] S2.2. In order to further characterize the probability of different SOCs during the charging process, the probability distribution of the initial and final charging SOCs is calculated using the cumulative probability distribution. The specific expression is as follows:

[0058]

[0059] Where f(SOC) represents the cumulative probability distribution function of starting or stopping charging SOC, and v represents the battery voltage;

[0060] According to the calculation results, the cumulative probability distribution curve of the voltage corresponding to the start-stop charging state of charge is drawn as follows: Figure 4 As shown;

[0061] Depend on Figure 4 It can be seen that the cumulative probability density increases with the increase of state of charge. This phenomenon may be attributed to range anxiety, which causes users to prefer charging the battery to full or nearly full.

[0062] S3. Extract the data of the battery charging process as health features during the cycle. The specific steps are as follows:

[0063] S3.1. As the battery age increases, polarization increases in the form of internal resistance, which is manifested as a reduction in constant current charging time. Therefore, the charging time or charging duration within a specific SOC range can effectively characterize battery aging, making it an easily measurable and valuable health feature. Therefore, this study uses the charging time within a specific SOC range as a health feature to characterize the battery capacity. However, different SOC ranges have different effects on capacity estimation. In order to minimize the capacity estimation error, the charging time within a specific SOC range is used as a health feature to characterize the battery capacity. In order to minimize the capacity estimation error, this study uses the Pearson correlation coefficient to quantify the relationship between health features and capacity. The specific expression is as follows:

[0064]

[0065] In the formula, x i represents the i-th feature, and They represent the average values ​​of characteristic sequence and capacity sequence respectively, and z represents capacity;

[0066] In addition, if Figure 2 , Figure 3 It can be seen that users show different probabilities of covering various SOC ranges during charging;

[0067] S3.2. Taking the cumulative probability density of the SOC range as the optimization target, the joint optimization objective function is obtained. The specific expression is as follows:

[0068]

[0069] In the formula, α and β are and the weight of the cumulative probability distribution, and β = 1-α, α is set to 0.5;

[0070] S3.3, in order to solve the objective function, the particle swarm optimization algorithm is used to minimize the objective function. The specific expression is as follows:

[0071]

[0072] Where i = 1, 2, ..., N, N is the total number of particles in the group, represents the position of particle i at the kth iteration, represents the velocity of particle i at the kth iteration, represents the speed of particle i at the k+1th iteration, ω is called the inertia factor, and its value is a non-negative number, c1 represents the individual optimal value learning factor, c2 represents the group optimal value learning factor, c1=c2=2, rand1() and rand2() represent random numbers uniformly distributed between [0, 1], rand1() may be equal to rand2() or not, represents the individual optimal position of particle i at the kth iteration, gbest k It represents the global optimal position of the entire particle swarm up to the kth iteration, and finally obtains the optimal SOC range, which is [4.083, 4.185];

[0073] Within the optimal SOC range, the probabilities of the initial charging voltage and the final charging voltage are 95.63% and 95.67% respectively, which indicates that when this SOC range is selected for health feature extraction, 95.63% of the charging process can normally extract health features for capacity estimation; in addition, within this range, the correlation between health features and capacity is 0.97, showing a strong correlation.

[0074] S4. Establish a remaining capacity estimation model based on health characteristics;

[0075] S4.1. Establish TCN model;

[0076] The temporal convolutional network (TCN) model is established, and the TCN model consists of causal convolution, dilated convolution and residual block. Causal convolution is used to prevent information leakage. The residual block has two dilated causal convolution layers, two ReLU layers, two normalization layers, and two Dropout layers. The one-dimensional full convolution network is implemented by padding the length to zero (kernel size is -1). The causal convolution operation at X is defined as:

[0077]

[0078] Where T is the size of the convolution kernel, t=1 means that the value of t starts from 1, t∈[1,T], x t is the value of the input time series at time t;

[0079] In addition to causal convolution, TCN also uses dilated convolution. t The dilated convolution operation at is defined as:

[0080]

[0081] Where X=(x1,x2,...,x t-1 ,x T ) is a one-dimensional sequence input, f=f0,f1,…,f k-1 Indicates filter, * d represents a dilated convolution operation with a dilation factor d, k is the filter size, k-1-id represents the past direction, and the receptive field is increased by increasing the dilation factor d. The value of the dilation factor is set to 2i;

[0082] S4.2, taking the charging time corresponding to the optimal SOC range as the input of the TCN neural network, and the corresponding remaining battery capacity as the output of the TCN network to complete the model construction;

[0083] The optimal SOC range is [4.083, 4.185];

[0084] S5. Estimate and verify the remaining capacity of the battery through the established model;

[0085] S5.1. Use the established model to estimate the remaining capacity of the battery and perform cross-validation;

[0086] In order to verify the proposed health feature optimization method and capacity estimation results, cross validation was performed using the Oxford University battery dataset;

[0087] In cross-validation, one cell is used as the test set and the other cells are used as the training set, and training and testing are performed iteratively;

[0088] This paper effectively evaluates the generalization ability of the proposed method. Figures 5 and 6 The capacity estimation results of different cells are given; it can be seen that the proposed capacity estimation method can effectively track the reference value, the maximum error occurs in battery 2, reaching 16.99mAh (2.3% of the nominal capacity), while the minimum estimation error is 7.43mAh (1% of the nominal capacity);

[0089] S5.2. Analyze the root mean square error (RMSE), maximum absolute error (MAE) and average absolute error (AAE) of each battery. The specific calculation formula is as follows:

[0090]

[0091]

[0092]

[0093] In the formula, y i represents the true value, Represents the predicted value, m represents the number of instances in the data set, and MAX() represents the maximum value;

[0094] The final calculation results are shown in Table 1. The maximum RMSE is 6.36mAh, the maximum MAE is 16.99mAh, and the maximum AAE is 5.03mAh, indicating that the proposed method can effectively estimate the battery capacity degradation with a maximum error of less than 2.3%.

[0095] Table 1: Capacity forecast error analysis

[0096]

[0097] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the remaining capacity of a lithium-ion battery based on a partial charging curve, characterized in that: The following steps are involved: S1. Collect data during the actual vehicle cycle charging process; S2. Analyze and count the collected charging data; S3, extracting data of the battery charging process as health features during the cycle; S4. Establish a remaining capacity estimation model based on health characteristics; S5. Estimate and verify the remaining battery capacity through the established model.

2. The method for predicting the remaining capacity of a lithium-ion battery based on a partial charging curve according to claim 1, characterized in that: The data of the actual vehicle cycle charging process is collected from randomly selected electric bicycles on the big data platform, including the battery charging voltage, charging time, charging start state of charge and charging end state of charge data.

3. The method for predicting the remaining capacity of a lithium-ion battery based on a partial charging curve according to claim 1, characterized in that: The specific steps of the analysis and statistics are as follows: S2.

1. Extract the frequency distribution of the initial and final charging state of charge from the collected data, establish a clustered bar chart with the battery state of charge as the horizontal axis and the number of occurrences as the vertical axis, and obtain the cumulative probability distribution curve of starting or stopping charging SOC; S2.

2. In order to further characterize the probability of different SOCs during the charging process, the probability distribution of the initial and final charging SOCs is calculated using the cumulative probability distribution. The specific expression is as follows: Where f(SOC) represents the cumulative probability distribution function of starting or stopping charging SOC, and v represents the battery voltage.

4. The method for predicting the remaining capacity of a lithium-ion battery based on a partial charging curve according to claim 1, characterized in that: The specific process of extracting the battery charging process data as health features during the cycle is as follows: S3.

1. The charging duration within the SOC range is used as a health feature to characterize the battery capacity. The Pearson correlation coefficient is used to quantify the relationship between the health feature and the capacity. The specific expression is as follows: In the formula, x i represents the i-th feature, and They represent the average values ​​of characteristic sequence and capacity sequence respectively, and z represents capacity; S3.

2. Taking the cumulative probability density of the SOC range as the optimization target, the joint optimization objective function is obtained. The specific expression is as follows: In the formula, α and β are and the weight of the cumulative probability distribution, and β = 1-α; S3.3, in order to solve the objective function, the particle swarm optimization algorithm is used to minimize the objective function. The specific expression is as follows: Where i = 1, 2, ..., N, N is the total number of particles in the group, represents the position of particle i at the kth iteration, represents the velocity of particle i at the kth iteration, represents the speed of particle i at the k+1th iteration, ω is called the inertia factor, and its value is a non-negative number, c1 represents the individual optimal value learning factor, c2 represents the group optimal value learning factor, c1=c2=2, rand1() and rand2() represent random numbers uniformly distributed between [0, 1], represents the individual optimal position of particle i at the kth iteration, gbest k It represents the global optimal position of the entire particle swarm up to the kth iteration, and finally obtains the optimal SOC range.

5. The method for predicting the remaining capacity of a lithium-ion battery based on a partial charging curve according to claim 1, characterized in that: The specific process of establishing the remaining capacity estimation model through health characteristics is as follows: S4.

1. Establish TCN model; S4.

2. The charging time corresponding to the optimal SOC range is used as the input of the TCN neural network, and the corresponding remaining battery capacity is used as the output of the TCN network to complete the model construction.

6. The method for predicting the remaining capacity of a lithium-ion battery based on a partial charging curve according to claim 1, characterized in that: The process of estimating and verifying the remaining battery capacity is as follows: S5.

1. Use the established model to estimate the remaining capacity of the battery and perform cross-validation; S5.

2. Analyze the root mean square error, maximum absolute error, and mean absolute error of each battery.

Images