Lithium battery SoC estimation method based on RFCNet model

Abstract

The invention relates to a lithium battery SoC (state of charge) estimation method based on an RFCNet (residual full-connection neural network) model, and the method is suitable for the estimation of the state of charge of a lithium ion battery. The method specifically comprises the following steps: acquiring basic data of the lithium battery; the method comprises the following steps: carrying out an experiment to construct an original data set, namely carrying out a plurality of charging and discharging experiments on a plurality of lithium ion batteries with the same model under different temperature conditions, recording experimental data at equal time intervals, and stopping the experiment until SoH (State of Health) is reduced to 70%; preprocessing the data and calculating SoH and SoC; dividing a data set: randomly dividing the data set into a training set and a test set according to a ratio of 7: 3; model training: training the RFCNet model by using the training set; and estimating the SoC of the current lithium battery by using the RFCNet model. According to the method, the SoC of the lithium battery is estimated based on the data such as the electrical parameters and the temperature and the RFCNet model, and convenience is provided for intelligent monitoring and lean management of the lithium battery.

Description

Technical Field

[0001] The present invention relates to the technical field of SoC estimation of storage elements, and in particular to a lithium battery SoC estimation method based on an RFCNet model. Background Art

[0002] Lithium-ion batteries are the main energy source for modern electric vehicles, portable devices and energy storage systems, and accurate estimation of their battery SoC (State of Charge) is crucial.

[0003] In the actual use of the battery, the operating conditions of the battery (such as current, ambient temperature, etc.) and the health status of the battery itself (such as battery aging, capacity attenuation, etc.) often change. These changing factors will affect the estimation of the battery power and cause the OCV-SoC curve provided by the battery when it leaves the factory to be no longer accurate. Although the traditional data-driven battery power estimation method takes into account key parameters such as battery voltage, current, and ambient temperature, it fails to comprehensively consider the relationship between ambient temperature, voltage, current, battery cell temperature, and battery SoH (battery health status State of Health) and the combined impact of the above parameters on SoC. It is often difficult to accurately reflect the actual state of the battery, which may lead to errors in the estimation of the battery charge state, thereby affecting the safe use of lithium batteries. Summary of the invention

[0004] In view of some shortcomings of the prior art, the present invention provides a lithium battery SoC estimation method based on the RFCNet model, which solves the problem of inaccurate state of charge estimation caused by incomplete data sampling or neglect of data correlation in existing related methods.

[0005] The technical solution adopted by the present invention to achieve the above-mentioned purpose is: a lithium battery SoC estimation method based on the RFCNet model, which specifically includes the following steps:

[0006] Step 1: Obtain basic data of lithium batteries for subsequent determination of experimental stop conditions and calculation of battery SoC;

[0007] Step 2: construct the original data set through experiments, that is, use multiple lithium-ion batteries of the same model to group them according to different ambient temperature conditions for cyclic charge and discharge experiments, record the experimental ambient temperature before the experiment, and record the battery cell temperature, voltage, current and corresponding time at equal time intervals during the experiment. Stop the experiment when SoH drops to 70%;

[0008] Step 3, preprocess the data and calculate SoH and SoC to obtain the battery experimental data set LD;

[0009] Step 4: Dataset division: randomly divide the dataset LD into a training set and a test set according to a ratio of 7:3;

[0010] Step 5: Model training: Use the divided training set to train the RFCNet model;

[0011] Step 6: Apply the trained RFCNet model to estimate the current lithium battery SoC.

[0012] As a further description of the above scheme, the basic data of the lithium battery obtained in step 1 specifically includes the rated capacity and OCV-SoC curve calibrated by the manufacturer at the factory;

[0013] As a further description of the above scheme, the experimental construction of the original data set in step 2 specifically includes the following steps:

[0014] S2.1. Select n lithium batteries of the same model and divide them into three groups. Perform charge-discharge cycle experiments at low temperature of 2-5℃, room temperature of 22-25℃, and high temperature of 42-45℃. Record the ambient temperature T before the experiment. Charge each group of lithium batteries at a constant charging current until the battery voltage reaches 4.2V. Then continue to charge at a constant voltage until the charging current drops to the charging cut-off current. Then, discharge at a constant discharge current until the battery cut-off voltage is reached and discharge stops.

[0015] S2.2. Repeat the charge-discharge process in S2.1, and record the battery voltage U, battery current I, and battery core temperature T at equal time intervals. C , when the SoH dropped to 70%, the experiment was stopped.

[0016] Preferably, in S2.1, the n lithium batteries of the same model are divided into three groups, and charge and discharge cycle experiments are carried out in room temperature 24° C., high temperature 44° C., and low temperature 4° C. environments, respectively.

[0017] As a further description of the above scheme, the data preprocessing in step 3 includes:

[0018] S3.1 Organize the data recorded in the experiment, check whether there are outliers and missing values ​​in the data, remove outliers in the data set, and use the mean interpolation method to fill in the missing values;

[0019] S3.2 Calculate the SoH of the battery after the i-th discharge as H i :

[0020]

[0021] Among them, H i represents the battery health status after the i-th experiment, Qi is the cumulative discharge amount of SoC from 100% to 0%, Q r is the rated capacity.

[0022] S3.3 Calculate the SoC during the i-th battery complete experiment based on the rated capacity of the battery as C t :

[0023]

[0024] Among them, t represents a certain moment in the experiment, C t It represents the state of charge of the lithium battery at time t, and I represents the current.

[0025] S3.4 obtains the battery experimental data set LD, LD = {T, U, I, T C , H i , C t}.

[0026] As a further description of the above scheme, in step 5, the model training described in step 5 includes:

[0027] The model is an RFCNet model, which is a fully connected structure. The model network structure is fixed and contains four hidden layers. The number of neurons in the first and second layers is 24, the number of neurons in the third and fourth layers is 12, and the number of neurons in the output layer is 1. A residual network is used to connect the first and third layers, and the second and fourth layers. The activation function used is ReLU. The specific formula of the model network structure is:

[0028] Z 1 =W1x+b1, formula (3)

[0029] Z 2 =W2Z 1 +b2, formula (4)

[0030] Z 3 =W3(Z 1 +Z 2 )+b3, formula (5)

[0031] Z 4 =W4(Z 2 +Z 3 )+b4, formula (6)

[0032] Z 5 =W5Z 4 +b5, formula (7)

[0033] a l =ReLU(Z l ), formula (8)

[0034] Where l represents the number of hidden layers (l = 1, 2, ..., 5), Z l represents the linear output of the lth layer, W l represents the weight of the lth layer, b l represents the bias of the lth layer, a l represents the activation output of the lth layer, and ReLU represents the activation function ReLU;

[0035] S5.2 uses the training set to train the RFCNet model. During the training process, the training parameters are adjusted by monitoring the changes in the training loss error. The loss function used during the training process is the absolute percentage error APE:

[0036]

[0037] Among them, y i represents the true value of SoC, Represents the estimated value of SoC. The training parameters that need to be adjusted include: optimizer, learning rate and batchsize to obtain the trained RFCNet model.

[0038] As a further description of the above scheme, in step 6, the ambient temperature T, battery voltage U, battery current I, battery core temperature T are input to the trained RFCNet model. C , battery health status SoH data, and obtain the SoC estimate of the lithium battery.

[0039] Preferably, in S5.2, the optimizer is selected from stochastic gradient descent SGD, average stochastic gradient descent ASGD, adaptive moment estimation Adam, and adaptive gradient algorithm AdaDelta; the learning rate is selected from 0.1, 0.01, and 0.001; and the batch size is selected from 10, 20, 30, and 50.

[0040] The present invention proposes a lithium battery SoC estimation method based on the RFCNet model, and the beneficial effects are specifically manifested in:

[0041] 1. The present invention selects ambient temperature, battery core temperature, voltage, current, and SoH during discharge in combination with lithium battery characteristics to estimate SoC. Compared with other methods, the present invention takes both electrical parameters and temperature parameters into account when estimating SoC, wherein the temperature parameter is used to compensate for changes in the internal characteristics of the battery, which will reduce the estimation error of the model;

[0042] 2. By adding residual connections to the fully connected network, we can better mine the key information contained in the ambient temperature, battery cell temperature, voltage, current, and SoH during discharge, as well as the correlation with SoC data, so as to more effectively obtain information in the data and effectively improve the model estimation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a block diagram of a lithium battery SoC estimation method based on the RFCNet model;

[0044] Figure 2 This is the network structure diagram of the RFCNet model;

[0045] Figure 3 This is the probability distribution diagram of the error of the lithium battery SoC estimation result; DETAILED DESCRIPTION

[0046] 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.

[0047] like Figure 1 As shown, a lithium battery SoC estimation method based on the RFCNet model includes the following steps:

[0048] Step 1: Obtain basic data of lithium batteries for subsequent determination of experimental stop conditions and calculation of battery SoC;

[0049] Step 2: Construct the original data set through experiments: Use multiple lithium-ion batteries of the same model to perform cyclic charge and discharge experiments according to different ambient temperature conditions. Record the ambient temperature before the experiment. Record the battery cell temperature, voltage, current and corresponding time at equal time intervals during the experiment. Stop the experiment when SoH drops to 70%.

[0050] Step 3, preprocess the data and calculate SoH and SoC to obtain the battery experimental data set LD;

[0051] Step 4: Dataset division: randomly divide the dataset LD into a training set and a test set according to a ratio of 7:3;

[0052] Step 5: Model training: Use the divided training set to train the RFCNet model;

[0053] Step 6: Apply the trained RFCNet model to estimate the current lithium battery SoC.

[0054] The basic data of the lithium battery obtained in step 1 of the present invention includes the rated capacity and OCV-SoC curve calibrated by the manufacturer before leaving the factory;

[0055] The step 2 of the present invention of conducting an experiment to construct an original data set includes the following steps:

[0056] S2.1. Select n lithium batteries of the same model and divide them into several groups. Perform charge-discharge cycle experiments in different temperature environments. Record the ambient temperature T before the experiment. Charge each group of lithium batteries at a constant charging current until the battery voltage reaches 4.2V. Then continue to charge at a constant voltage until the charging current drops to the charging cut-off current. Then, discharge at a constant discharge current until the battery cut-off voltage is reached and discharge stops.

[0057] S2.2. Repeat the charge-discharge process in S2.1, and record the battery voltage U, battery current I, and battery core temperature T at equal time intervals. C ,When the SoH drops to 70%, the experiment is stopped;

[0058] Taking the first battery as an example, the battery needs to be discharged q times at a constant current in a constant temperature environment. The ambient temperature T is recorded before the experiment begins, and the battery voltage U, battery current I, and battery core temperature T are recorded at equal time intervals after the battery starts discharging. C , charge at a constant current of 1.5A until the battery voltage reaches 4.2V, then continue charging at a constant voltage until the charging current drops to 20mA and stop charging, discharge at a constant current of 2A until the battery cut-off voltage is reached and stop discharging, and stop the experiment when SoH drops to 70%;

[0059] The data preprocessing in step 3 of the present invention includes the following steps:

[0060] Check whether there are outliers and missing values ​​in the data set, remove outliers in the data set, and use mean interpolation method to fill in missing values;

[0061] Calculate the SoH of the battery after the i-th discharge:

[0062]

[0063] Among them, H i represents the battery health status after the i-th discharge (i=1,2,…,q), Q i is the cumulative discharge amount of SoC from 100% to 0%, Q r is the rated capacity.

[0064] Calculate the SoC during the i-th battery experiment based on the battery rated capacity:

[0065]

[0066] Among them, t represents a certain moment in the experiment, C t It represents the state of charge of the lithium battery at time t, and I represents the current.

[0067] Get the battery experimental data set LD, LD = {T, U, I, T C , H i , C t}.

[0068] The model training in step 5 of the present invention includes the following steps:

[0069] The model is an RFCNet model, which is a fully connected structure. The network structure is fixed and contains four hidden layers. The number of neurons in the first and second layers is 24, the number of neurons in the third and fourth layers is 12, and the number of neurons in the output layer is 1. A residual network is used to connect the first and third layers, and the second and fourth layers. The activation function used is ReLU. The specific formula of the model network structure is:

[0070] Z 1 =W1x+b1

[0071] Z 2 =W2Z 1 +b2

[0072] Z 3 =W3(Z 1 +Z 2 )+b3

[0073] Z 4 =W4(Z 2 +Z 3 )+b4

[0074] Z 5 =W5Z 4 +b5

[0075] a l =ReLU(Z l )

[0076] Where l represents the number of hidden layers (l = 1, 2, ..., 5), Z l represents the linear output of the lth layer, W l represents the weight of the lth layer, b l represents the bias of the lth layer, a l represents the activation output of the lth layer, and ReLU represents the activation function ReLU;

[0077] The RFCNet model is trained with the training set. During the training process, the training parameters are adjusted by monitoring the changes in the training loss error. The loss function used in the training process is APE:

[0078]

[0079] Where yi represents the true value of SoC, Represents the estimated value of SoC. The training parameters that need to be adjusted include: optimizer, learning rate and batchsize (where the optimizer is selected from SGD (stochastic gradient descent), ASGD (averaged stochastic gradient descent), Adam (adaptive moment estimation), AdaDelta (adaptive gradient algorithm), the learning rate is selected from 0.1, 0.01, 0.001, and the batchsize is selected from 10, 20, 30, 50), and the RFCNet model is obtained.

[0080] like Figure 1 As shown, the lithium battery SoC estimation method based on the RFCNet model of an embodiment of the present invention has the following specific operation process:

[0081] Obtain basic data on lithium batteries; conduct experiments to construct the original data set, use multiple lithium-ion batteries of the same model to conduct multiple charge and discharge experiments under different temperature conditions, and record the experimental environment temperature T, voltage U, current I, and battery core temperature T during the experiment at equal time intervals. c and the corresponding time t, and the experiment is stopped when SoH drops to 70%; the data is preprocessed and SoH and SoC are calculated to obtain the battery experiment data set LD; the data set LD is randomly divided into a training set and a test set according to a ratio of 7:3; the RFCNet model is trained using the divided training set to obtain the RFCNet model; the RFCNet model is applied to estimate the current lithium battery SoC.

[0082] In order to explain the above solution more intuitively, in the following content, the above solution is described in detail in combination with specific embodiments.

[0083] Step 1: This scheme selects 12 18650 lithium batteries for experiment and obtains the basic data of the 12 lithium batteries, including the rated capacity of 2Ah and OCV-SoC curve calibrated by the manufacturer at the factory;

[0084] Step 2: Conduct experiments to construct raw data. Number the 12 18650 lithium-ion batteries as 1, 2, 3, 4, ..., 11, and 12 batteries, and conduct multiple charge and discharge experiments under different temperature conditions. Among them, batteries 1, 2, 3, and 4 were subjected to multiple charge and discharge experiments at room temperature of 24°C, batteries 5, 6, 7, and 8 were subjected to high temperature of 44°C, and batteries 9, 10, 11, and 12 were subjected to low temperature of 4°C. During the experiment, discharge was performed at a constant current of 2A until the voltage dropped to the cut-off voltage of 2.7V and the current discharge was stopped. The ambient temperature T was recorded before the experiment began, and the voltage U, current I, and battery core temperature T at the corresponding time intervals during the experiment were recorded. C , the rated capacity was reduced from 2Ah to 1.4Ah and the experiment was stopped.

[0085] Step 3: Preprocess the experimental data, organize the experimental record data, check whether there are outliers and missing values ​​in the data, remove the outliers in the data set, and fill the missing values ​​using the average of the previous and next data;

[0086] Calculate the SoH of the battery after the i-th discharge according to the maximum discharge capacity and rated capacity during each experiment:

[0087]

[0088] Among them, H i represents the battery health status after the i-th discharge (i=1,2,…,q), Q i is the cumulative discharge amount of SoC from 100% to 0%, Q r is the rated capacity.

[0089] Calculate the SoC during the i-th battery complete experiment based on the battery rated capacity:

[0090]

[0091] Among them, t represents a certain moment in the experiment, C t It represents the state of charge of the lithium battery at time t, and I represents the current.

[0092] The data obtained in this scheme include the ambient temperature before discharge, SoH, voltage, current, battery core temperature, and SoC during the experiment, and the battery experiment data set LD is obtained, LD = {T, U, I, T C , H i , C t}.

[0093] Step 4: Divide the data set into training set and test set randomly according to the ratio of 7:3.

[0094] Step 5: Perform model training, such as Figure 2 As shown, the model is a fully connected structure with a fixed network structure, which contains four hidden layers. The number of neurons in the first and second layers is 24, the number of neurons in the third and fourth layers is 12, and the number of neurons in the output layer is 1. The residual network (Residual Network, ResNet) is used to connect the first and third layers, the second and fourth layers, and the activation function used is ReLU. The specific formula of the RFCNet model network structure is:

[0095] Z 1 =W1x+b1, formula (3)

[0096] Z 2 =W2Z 1 +b2, formula (4)

[0097] Z 3 =W3(Z 1 +Z 2 )+b3, formula (5)

[0098] Z 4 =W4(Z 2 +Z 3 )+b4, formula (6)

[0099] Z 5 =W5Z 4 +b5, formula (7)

[0100] a l =ReLU(Z l ), formula (8)

[0101] Where l represents the number of hidden layers (l = 1, 2, ..., 5), Z l represents the linear output of the lth layer, W l represents the weight of the lth layer, b l represents the bias of the lth layer, a l Represents the activation output of the lth layer, and ReLU represents the activation function ReLU.

[0102] The RFCNet model is trained with the training set. During the training process, the training parameters are adjusted by monitoring the changes in the training loss error. The loss function used in the training process is APE:

[0103]

[0104] Among them, y i represents the true value of SoC, Represents the estimated value of SoC. The training parameters that need to be adjusted include: optimizer, learning rate and batchsize (the optimizer is selected from SGD (stochastic gradient descent), ASGD (average stochastic gradient descent), Adam (adaptive moment estimation), AdaDelta (adaptive gradient algorithm), the learning rate is selected from 0.1, 0.01, 0.001, and the batchsize is selected from 10, 20, 30, 50). During the experiment, the loss function curves under different training parameter combinations were observed and the final training parameter combination was determined as follows: the optimizer uses Adam, the learning rate is 0.01, and the batchsize is 30. The RFCNet model is obtained by training with the training set under this training parameter combination.

[0105] Step 6: Apply the RFCNet model to estimate the current lithium battery SoC.

[0106] The test set containing 14214 data items was used for testing. The error probability distribution diagram of the lithium battery SoC estimation result is shown in the figure below. Figure 3 As shown. The horizontal axis in the figure represents the absolute error of each lithium battery SoC estimation result, and the vertical axis represents the probability density distribution of the error. It can be clearly seen that the error distribution of the RFCNet model proposed in this patent on the test set presents a unimodal feature, with a peak value of about 0.8%, an average value of 1.5%, and most of the errors are small. These experimental results show that the model can accurately estimate the SoC of the battery during the experiment, and the overall prediction effect is good.

Claims

1. A lithium battery SoC estimation method based on the RFCNet model, characterized in that: The specific steps include: Step 1: Obtain basic data of lithium batteries for subsequent determination of experimental stop conditions and calculation of battery SoC; Step 2: Construct the original data set through experiments: Use multiple lithium-ion batteries of the same model to perform cyclic charge and discharge experiments according to different ambient temperature conditions. Record the ambient temperature before the experiment. Record the battery cell temperature, voltage, current and corresponding time at equal time intervals during the experiment. Stop the experiment when SoH drops to 70%. Step 3, preprocess the data and calculate SoH and SoC to obtain the battery experimental data set LD; Step 4: Dataset division: randomly divide the dataset LD into a training set and a test set according to a ratio of 7:3; Step 5: Model training: Use the divided training set to train the RFCNet model; Step 6: Apply the trained RFCNet model to estimate the current lithium battery SoC.

2. According to a lithium battery SoC estimation method based on the RFCNet model according to claim 1, it is characterized in that: The basic data of the lithium battery described in step 1 specifically includes the rated capacity and OCV-SoC curve calibrated by the manufacturer before leaving the factory.

3. The lithium battery SoC estimation method based on the RFCNet model according to claim 1, characterized in that: The experiment described in step 2 to construct the original data set specifically includes the following steps: S2.

1. Select n lithium batteries of the same model and divide them into three groups. Perform charge-discharge cycle experiments at low temperature of 2-5℃, room temperature of 22-25℃, and high temperature of 42-45℃. Record the ambient temperature T before the experiment. Charge each group of lithium batteries at a constant charging current until the battery voltage reaches 4.2V. Then continue to charge at a constant voltage until the charging current drops to the charging cut-off current. Then, discharge at a constant discharge current until the battery cut-off voltage is reached and discharge stops. S2.

2. Repeat the charge-discharge process in S2.1, and record the battery voltage U, battery current I, and battery core temperature T at equal time intervals. C , when the SoH dropped to 70%, the experiment was stopped.

4. The lithium battery SoC estimation method based on the RFCNet model according to claim 1, characterized in that: The data preprocessing and calculation of SoH and SoC described in step 3 include: S3.1 Organize the data recorded in the experiment, check whether there are outliers and missing values ​​in the data, remove outliers in the data set, and use the mean interpolation method to fill in the missing values; S3.2 Calculate the SoH of the battery after the i-th discharge as H i : Among them, H i represents the battery health status after the i-th experiment, Q i is the cumulative discharge amount of SoC from 100% to 0%, Q r is the rated capacity; S3.3 Calculate the SoC during the i-th battery complete experiment based on the rated capacity of the battery as C t : Among them, t represents a certain moment in the experiment, C t It represents the state of charge of the lithium battery at time t, and I represents the current; S3.4 obtains the battery experimental data set LD, LD = {T, U, I, T C , H i , C t }.

5. The lithium battery SoC estimation method based on the RFCNet model according to claim 1, characterized in that: The model training described in step 5 includes: The model described in S5.1 is the RFCNet model, which is a fully connected structure. The model network structure is fixed and contains four hidden layers. The number of neurons in the first and second layers is 24, the number of neurons in the third and fourth layers is 12, and the number of neurons in the output layer is 1. A residual network is used to connect the first and third layers, and the second and fourth layers. The activation function used is ReLU. The specific formula of the model network structure is: Z 1 = W1x + b1, formula (3) Z 2 =W2Z 1 +b2, formula (4) Z 3 = W3(Z 1 + Z 2 ) + b3, Equation (5) Z 4 =W4(Z 2 +Z 3 )+b4, formula (6) Z 5 = W5Z 4 +b5, formula (7) a l =ReLU(Z l ), formula (8) Where l represents the number of hidden layers (l = 1, 2, ..., 5), Z l represents the linear output of the lth layer, W l represents the weight of the lth layer, b l represents the bias of the lth layer, a l represents the activation output of the lth layer, and ReLU represents the activation function ReLU; S5.2 uses the training set to train the RFCNet model. During the training process, the training parameters are adjusted by monitoring the changes in the training loss error. The loss function used during the training process is the absolute percentage error APE: Among them, y i represents the true value of SoC, Represents the estimated value of SoC. The training parameters that need to be adjusted include: optimizer, learning rate and batchsize to obtain the trained RFCNet model.

6. The lithium battery SoC estimation method based on the RFCNet model according to claim 1, characterized in that: In step 6, the trained RFCNet model is fed with the ambient temperature T, battery voltage U, battery current I, and battery core temperature T. C , battery health status SoH data, and obtain the SoC estimate of the lithium battery.

7. The lithium battery SoC estimation method based on the RFCNet model according to claim 3, characterized in that: In S2.1, the n lithium batteries of the same model are divided into three groups, and charge and discharge cycle experiments are performed in room temperature 24° C., high temperature 44° C., and low temperature 4° C. environments, respectively.

8. The lithium battery SoC estimation method based on the RFCNet model according to claim 5, characterized in that: In S5.2, the optimizer is selected from stochastic gradient descent SGD, average stochastic gradient descent ASGD, adaptive moment estimation Adam, and adaptive gradient algorithm AdaDelta; the learning rate is selected from 0.1, 0.01, and 0.001; and the batch size is selected from 10, 20, 30, and 50.

Images