A method for estimating the state of health of a lithium-ion battery of an energy storage power station

Abstract

The present application relates to a kind of SOH estimation methods of energy storage power station lithium ion battery, the method includes: the relaxation voltage data and SOH value after lithium ion battery full charge are collected;From the extracted feature of acquisition data and calculate first-order difference voltage data;Data set is divided into training set and test set;LightGBM model is constructed, and training set and test set are used to model training and test respectively;One-dimensional convolutional neural network CNN is constructed, and training set and test set are used to model training and test respectively;The extracted feature, the SOH value estimated by LightGBM model and CNN model is combined into feature data set;Linear regression LR model is constructed, and it is trained and tested using feature data set;The extracted feature, the SOH value estimated by LightGBM model and CNN model is input into LR model, and the final SOH estimation result is obtained.The method has good generalization ability, and estimation precision is high.

Description

Technical Field

[0001] This invention relates to the field of battery energy storage technology, and specifically to a method for estimating the state of harm (SOH) of lithium-ion batteries in energy storage power stations. Background Technology

[0002] With the depletion of traditional energy sources and the deterioration of the ecological environment in recent years, the research and development of new energy sources has received widespread attention. Lithium-ion batteries, as one type of new energy, are gradually being widely used in grid energy storage, electric vehicles, and aerospace. Over time, the State of Health (SOH) of lithium-ion batteries undergoes an irreversible decline. When the SOH falls below a certain threshold, the battery's energy storage performance deteriorates, and the failure rate increases. Therefore, accurately estimating the SOH of a battery is crucial for its safe use.

[0003] Existing SOH estimation methods for lithium-ion batteries impose strict requirements on the battery's operating conditions. For example, methods based on partial constant-current charging curves require the battery to fully pass through a predetermined voltage window during charging; methods based on constant-voltage charging curves are affected by the charging current; and methods based on discharge voltage curves are difficult to apply in real-world operating scenarios. Furthermore, data-driven SOH estimation methods only consider manual or automatic feature extraction using deep learning models. However, both feature extraction methods have their own characteristics and advantages and are not mutually exclusive. Therefore, there is a need to develop an SOH estimation method that has lower requirements for battery operating conditions and integrates both manual and automatic feature extraction. Summary of the Invention

[0004] The purpose of this invention is to provide a method for estimating the state of oxygen (SOH) of lithium-ion batteries in energy storage power stations. This method has good generalization ability and high estimation accuracy.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is: a method for estimating the state of harm (SOH) of lithium-ion batteries in energy storage power stations, comprising the following steps:

[0006] S1: Collect the relaxation voltage sequence data V of the lithium-ion battery during the 5-minute rest period after full charge in each cycle, and the SOH label value of the battery in each cycle to form the original dataset D1;

[0007] S2: Based on the original dataset D1 obtained in step S1, calculate the maximum value V of the relaxation voltage sequence data in each cycle. max Minimum value V min Shannon entropy V entropy Average value V mean and standard deviation V std As features, they, along with the SOH label values, form the feature dataset D. f,1Calculate the first-order differential voltage sequence data ∆V of the relaxed voltage sequence data, and combine the relaxed voltage sequence data V, the first-order differential voltage sequence data ∆V, and the corresponding SOH label values ​​to form a feature dataset D. f,2 ;

[0008] S3: Divide the feature dataset D using the same partitioning method. f,1 and D f,2 The dataset is divided into training sets D. train,1 and D train,2 Test set D test,1 and D test,2 ;

[0009] S4: Construct the LightGBM model, using the training set D respectively. train,1 and test set D test,1 The LightGBM model is trained and tested. The mean absolute error (MAE) is used to measure the estimation accuracy of the LightGBM model. When the MAE of the model on the test set is ≤2%, the model is considered to have completed training. Otherwise, step S4 is repeated until the condition is met.

[0010] S5: Construct a one-dimensional convolutional neural network (CNN), using the training set D respectively. train,2 and test set D test,2 The CNN model is trained and tested. The mean absolute error (MAE) is used to measure the estimation accuracy of the CNN model. When the MAE of the model on the test set is ≤2%, the model is considered to have completed training. Otherwise, step S5 is repeated until the condition is met.

[0011] S6: Extract the 5 relaxation voltage features from step S2, i.e., the maximum value V max Minimum value V min Shannon entropy V entropy Average value V mean and standard deviation V std And the LightGBM model on the feature dataset D f,1 The SOH estimation results output above, and the CNN model on the feature dataset D f,2 The SOH estimation results output above are merged and combined with the SOH label values ​​to form the feature dataset D. f,3 and the feature dataset D f,3 The dataset is partitioned according to the partitioning method in step S3 to obtain the training set D. train,3 and test set D test,3 ;

[0012] S7: Construct the LR model, using the training set D respectively. train,3 and test set D test,3The LR model is trained and tested. Similarly, the mean absolute error (MAE) is used to measure the estimation accuracy of the linear regression LR model. When the MAE of the model on the test set is ≤1.5%, the model is considered to have completed training. Otherwise, step S7 is repeated until the condition is met.

[0013] S8: Import the LightGBM, CNN, and LR models trained in steps S4, S5, and S7 into the battery management system. After the lithium-ion battery in the energy storage station has finished charging and has been left to stand for 5 minutes, extract features from the collected relaxation voltage sequence data and calculate the first-order differential voltage sequence data in the manner of step S2. Input the 5 relaxation voltage features into the LightGBM model to obtain the SOH estimation result. Input the original relaxation voltage sequence data and the first-order differential voltage sequence data into the CNN model to obtain the SOH estimation result. Then, input the SOH estimation results output by the two models and the 5 relaxation voltage features into the LR model to obtain the final SOH estimation result.

[0014] Furthermore, step S1 specifically includes the following steps:

[0015] S1-1: Collect the relaxation voltage sequence data V of the lithium-ion battery during the 5-minute rest period after full charging in each cycle. The specific format of the sequence data for the i-th cycle is as follows:

[0016]

[0017] in, This represents the nth voltage value in the i-th cycle;

[0018] S1-2: The relaxation voltage sequence data V and the SOH label value of the battery in each cycle are combined to form the original dataset D1, which is as follows:

[0019] .

[0020] Furthermore, step S2 specifically includes the following steps:

[0021] S2-1: Based on the original dataset D1 obtained in step S1, calculate the maximum value V of the relaxation voltage sequence data in each cycle. max Minimum value V min Shannon entropy V entropy Average value V mean and standard deviation V std As features, they, along with the SOH label values, form the feature dataset D. f,1 And after normalization, its specific form is as follows:

[0022]

[0023] in, This represents the maximum relaxation voltage in the i-th cycle. This represents the minimum relaxation voltage in the i-th cycle. The Shannon entropy value of the relaxation voltage in the i-th cycle is represented by the following formula: , This represents the probability of the nth element appearing in a data sequence, where N represents the total number of values ​​in the data sequence. This represents the average relaxation voltage during the i-th cycle. This represents the standard deviation of the relaxation voltage in the i-th cycle;

[0024] S2-2: Calculate the first-order differential voltage sequence data ∆V of the relaxed voltage sequence data. The specific form of the sequence data in the i-th cycle is as follows:

[0025]

[0026] in, The relaxation voltage sequence data V, the first-order differential voltage sequence data ∆V, and the corresponding SOH label values ​​are combined to form a feature dataset D. f,2 And after normalization, its specific form is as follows:

[0027] .

[0028] Furthermore, step S4 specifically includes the following steps:

[0029] S4-1: Build a LightGBM model and set the model's learning rate, number of estimators, and maximum tree depth;

[0030] S4-2: Using the training set D obtained in step S3 train,1 and test set D test,1 The LightGBM model was trained and tested, and the mean absolute error (MAE) was used to measure the estimation accuracy of the LightGBM model. The formula for calculating MAE is as follows:

[0031]

[0032] in, This represents the SOH estimate of the battery in the l-th cycle, as output by the model. This represents the actual SOH value of the battery in the lth cycle;

[0033] The LightGBM model completes training when its MAE on the test set is ≤2%; otherwise, repeat step S4 until this condition is met.

[0034] Furthermore, step S5 specifically includes the following steps:

[0035] S5-1: Construct a one-dimensional convolutional neural network (CNN), and set the number of layers, number of convolutional kernels, kernel size, kernel stride, and learning rate.

[0036] S5-2: Using the training set D obtained in step S3 train,2 and test set D test,2 The CNN model is trained and tested. The mean absolute error (MAE) is used to measure the estimation accuracy of the CNN model. When the MAE of the model on the test set is ≤2%, the model is considered to have completed training. Otherwise, step S5 is repeated until the condition is met.

[0037] Furthermore, step S6 specifically includes the following steps:

[0038] S6-1: Extract the 5 relaxation voltage features from step S2, i.e., the maximum value V max Minimum value V min Shannon entropy V entropy Average value V mean and standard deviation V std And the LightGBM model on the feature dataset D f,1 The SOH estimation results output above, and the CNN model on the feature dataset D f,2 The SOH estimation results output above are merged and combined with the SOH label values ​​to form the feature dataset D. f,3 Its specific form is as follows:

[0039]

[0040] in, This represents the SOH estimate output by the LightGBM model in the i-th loop. This represents the SOH estimate output by the CNN model in the i-th loop;

[0041] S6-2: Transfer the feature dataset D f,3 The dataset is partitioned according to the partitioning method in step S3 to obtain the training set D. train,3 and test set D test,3 .

[0042] Furthermore, step S7 specifically includes the following steps:

[0043] S7-1: Construct a linear regression (LR) model;

[0044] S7-2: Using the training set D obtained in step S6 train,3 and test set D test,3The LR model is trained and tested. The mean absolute error (MAE) is used to measure the estimation accuracy of the LR model. When the MAE of the LR model on the test set is ≤1.5%, the model is considered to have completed training. Otherwise, step S7-2 is repeated until the condition is met.

[0045] Compared with existing technologies, this invention has the following advantages: It provides a method for estimating the state of charge (SOH) of lithium-ion batteries in energy storage power stations. This method estimates SOH based on relaxation voltage data during a short resting period after the battery is fully charged. It has no specific requirements on the battery's charging and discharging current and discharge mode, effectively reducing the limitations imposed by this method on battery operating conditions. Furthermore, by fusing five manually extracted relaxation voltage features, the SOH estimated from these five features, and the SOH estimated from features automatically extracted by a CNN model using an LR model, it effectively integrates the advantages of both manual and automatic feature extraction, and further utilizes the manually extracted features. Therefore, this method effectively improves the estimation accuracy. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the method implementation of an embodiment of the present invention. Detailed Implementation

[0047] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0048] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0049] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0050] like Figure 1 As shown in the figure, this embodiment provides a method for estimating the state of harm (SOH) of a lithium-ion battery in an energy storage power station, including the following steps:

[0051] S1: Collect the relaxation voltage sequence data V of the lithium-ion battery during the 5-minute rest period after full charge in each cycle, and the SOH label value of the battery in each cycle to form the original dataset D1.

[0052] S2: Based on the original dataset D1 obtained in step S1, calculate the maximum value V of the relaxation voltage sequence data in each cycle.max Minimum value V min Shannon entropy V entropy Average value V mean and standard deviation V std As features, they, along with the SOH label values, form the feature dataset D. f,1 Calculate the first-order differential voltage sequence data ∆V of the relaxed voltage sequence data, and combine the relaxed voltage sequence data V, the first-order differential voltage sequence data ∆V, and the corresponding SOH label values ​​to form a feature dataset D. f,2 .

[0053] S3: Divide the feature dataset D using the same partitioning method. f,1 and D f,2 The dataset is divided into training sets D. train,1 and D train,2 Test set D test,1 and D test,2 .

[0054] S4: Construct the LightGBM model, using the training set D respectively. train,1 and test set D test,1 The LightGBM model is trained and tested. The mean absolute error (MAE) is used to measure the estimation accuracy of the LightGBM model. When the MAE of the model on the test set is ≤2%, the model is considered to have completed training. Otherwise, step S4 is repeated until the condition is met.

[0055] S5: Construct a one-dimensional convolutional neural network (CNN), using the training set D respectively. train,2 and test set D test,2 The CNN model is trained and tested. Similarly, the mean absolute error (MAE) is used to measure the estimation accuracy of the CNN model. When the MAE of the model on the test set is ≤2%, the model is considered to have completed training. Otherwise, step S5 is repeated until the condition is met.

[0056] S6: Extract the 5 relaxation voltage features from step S2, i.e., the maximum value V max Minimum value V min Shannon entropy V entropy Average value V mean and standard deviation V std And the LightGBM model on the feature dataset D f,1 The SOH estimation results output above, and the CNN model on the feature dataset D f,2 The SOH estimation results output above are merged and combined with the SOH label values ​​to form the feature dataset D. f,3 and the feature dataset D f,3 The dataset is partitioned according to the partitioning method in step S3 to obtain the training set D. train,3 and test set Dtest,3 .

[0057] S7: Construct a linear regression (LR) model, using the training set D respectively. train,3 and test set D test,3 The LR model is trained and tested. Similarly, the mean absolute error (MAE) is used to measure the estimation accuracy of the linear regression LR model. When the MAE of the model on the test set is ≤1.5%, the model is considered to have completed training. Otherwise, step S7 is repeated until this condition is met.

[0058] S8: Import the LightGBM, CNN, and LR models trained in steps S4, S5, and S7 into the battery management system. After the lithium-ion battery in the energy storage station has finished charging and has been left to stand for 5 minutes, extract features from the collected relaxation voltage sequence data and calculate the first-order differential voltage sequence data in the manner of step S2. Input the 5 relaxation voltage features into the LightGBM model to obtain the SOH estimation result. Input the original relaxation voltage sequence data and the first-order differential voltage sequence data into the CNN model to obtain the SOH estimation result. Then, input the SOH estimation results output by the two models and the 5 relaxation voltage features into the LR model to obtain the final SOH estimation result.

[0059] In this embodiment, step S1 specifically includes the following steps:

[0060] S1-1: Collect the relaxation voltage sequence data V of the lithium-ion battery during the 5-minute rest period after full charge in each cycle. Taking the sequence data of the i-th cycle as an example, its specific form is as follows:

[0061]

[0062] in, This represents the nth voltage value in the i-th cycle.

[0063] S1-2: The relaxation voltage sequence data V and the SOH label value of the battery in each cycle are combined to form the original dataset D1, which is as follows:

[0064] .

[0065] Step S2 specifically includes the following steps:

[0066] S2-1: Based on the original dataset D1 obtained in step S1, calculate the maximum value V of the relaxation voltage sequence data in each cycle. max Minimum value V min Shannon entropy V entropy Average value V mean and standard deviation V std As features, they, along with the SOH label values, form the feature dataset D.f,1 And after normalization, its specific form is as follows:

[0067]

[0068] in, This represents the maximum relaxation voltage in the i-th cycle. This represents the minimum relaxation voltage in the i-th cycle. The Shannon entropy value of the relaxation voltage in the i-th cycle is represented by the following formula: , This represents the probability of the nth element appearing in a data sequence, where N represents the total number of values ​​in the data sequence. This represents the average relaxation voltage during the i-th cycle. This represents the standard deviation of the relaxation voltage in the i-th cycle.

[0069] S2-2: Calculate the first-order differential voltage sequence data ∆V of the relaxed voltage sequence data. Taking the i-th loop as an example, its specific form is as follows:

[0070]

[0071] in, The relaxation voltage sequence data V, the first-order differential voltage sequence data ∆V, and the corresponding SOH label values ​​are combined to form a feature dataset D. f,2 And after normalization, its specific form is as follows:

[0072] .

[0073] Step S4 specifically includes the following steps:

[0074] S4-1: Build the LightGBM model and set parameters such as the learning rate, number of estimators, and maximum tree depth.

[0075] S4-2: Using the training set D obtained in step S3 train,1 and test set D test,1 The LightGBM model was trained and tested, and the mean absolute error (MAE) was used to measure the estimation accuracy of the LightGBM model. The formula for calculating MAE is as follows:

[0076]

[0077] in, This represents the SOH estimate of the battery in the l-th cycle, as output by the model. This represents the actual SOH value of the battery in the lth cycle.

[0078] The LightGBM model completes training when its MAE on the test set is ≤2%; otherwise, repeat step S4 until this condition is met.

[0079] Step S5 specifically includes the following steps:

[0080] S5-1: Construct a one-dimensional convolutional neural network (CNN), setting the number of layers, number of convolutional kernels, kernel size, kernel stride, and learning rate.

[0081] S5-2: Using the training set D obtained in step S3 train,2 and test set D test,2 The CNN model is trained and tested. Similarly, the mean absolute error (MAE) is used to measure the estimation accuracy of the CNN model. When the MAE of the model on the test set is ≤2%, the model is considered to have completed training. Otherwise, step S5 is repeated until the condition is met.

[0082] Step S6 specifically includes the following steps:

[0083] S6-1: Extract the 5 relaxation voltage features from step S2, i.e., the maximum value V max Minimum value V min Shannon entropy V entropy Average value V mean and standard deviation V std And the LightGBM model on the feature dataset D f,1 The SOH estimation results output above, and the CNN model on the feature dataset D f,2 The SOH estimation results output above are merged and combined with the SOH label values ​​to form the feature dataset D. f,3 Its specific form is as follows:

[0084]

[0085] in, This represents the SOH estimate output by the LightGBM model in the i-th loop. This represents the estimated SOH value output by the CNN model in the i-th loop.

[0086] S6-2: Transfer the feature dataset D f,3 The dataset is partitioned according to the partitioning method in step S3 to obtain the training set D. train,3 and test set D test,3 .

[0087] Step S7 specifically includes the following steps:

[0088] S7-1: Construct a linear regression (LR) model;

[0089] S7-2: Using the training set D obtained in step S6 train,3 and test set D test,3 The LR model is trained and tested. The mean absolute error (MAE) is used to measure the estimation accuracy of the LR model. When the MAE of the LR model on the test set is ≤1.5%, the model is considered to have completed training. Otherwise, step S7-2 is repeated until the condition is met.

[0090] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0091] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0092] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0093] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0094] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for estimating the state of harm (SOH) of a lithium-ion battery in an energy storage power station, characterized in that, Includes the following steps: S1: Collect the relaxation voltage sequence data V of the lithium-ion battery during the 5-minute rest period after full charge in each cycle, and the SOH label value of the battery in each cycle to form the original dataset D1; S2: Based on the original dataset D1 obtained in step S1, calculate the maximum value V of the relaxation voltage sequence data in each cycle. max Minimum value V min Shannon entropy V entropy Average value V mean and standard deviation V std As features, they, along with the SOH label values, form the feature dataset D. f,1 Calculate the first-order differential voltage sequence data ∆V of the relaxed voltage sequence data, and combine the relaxed voltage sequence data V, the first-order differential voltage sequence data ∆V, and the corresponding SOH label values ​​to form a feature dataset D. f,2 ; S3: Divide the feature dataset D using the same partitioning method. f,1 and D f,2 The dataset is divided into training sets D. train,1 and D train,2 Test set D test,1 and D test,2 ; S4: Construct the LightGBM model, using the training set D respectively. train,1 and test set D test,1 The LightGBM model is trained and tested. The mean absolute error (MAE) is used to measure the estimation accuracy of the LightGBM model. When the MAE of the model on the test set is ≤2%, the model is considered to have completed training. Otherwise, step S4 is repeated until the condition is met. S5: Construct a one-dimensional convolutional neural network (CNN), using the training set D respectively. train,2 and test set D test,2 The CNN model is trained and tested. The mean absolute error (MAE) is used to measure the estimation accuracy of the CNN model. When the MAE of the model on the test set is ≤2%, the model is considered to have completed training. Otherwise, step S5 is repeated until the condition is met. S6: Extract the 5 relaxation voltage features from step S2, i.e., the maximum value V max Minimum value V min Shannon entropy V entropy Average value V mean and standard deviation V std And the LightGBM model on the feature dataset D f,1 The SOH estimation results output above, and the CNN model on the feature dataset D f,2 The SOH estimation results output above are merged and combined with the SOH label values ​​to form the feature dataset D. f,3 and the feature dataset D f,3 The dataset is partitioned according to the partitioning method in step S3 to obtain the training set D. train,3 and test set D test,3 ; S7: Construct the LR model, using the training set D respectively. train,3 and test set D test,3 The LR model is trained and tested. Similarly, the mean absolute error (MAE) is used to measure the estimation accuracy of the linear regression LR model. When the MAE of the model on the test set is ≤1.5%, the model is considered to have completed training. Otherwise, step S7 is repeated until the condition is met. S8: Import the LightGBM, CNN, and LR models trained in steps S4, S5, and S7 into the battery management system. After the lithium-ion battery in the energy storage station has finished charging and has been left to stand for 5 minutes, extract features from the collected relaxation voltage sequence data and calculate the first-order differential voltage sequence data in the manner of step S2. Input the 5 relaxation voltage features into the LightGBM model to obtain the SOH estimation result. Input the original relaxation voltage sequence data and the first-order differential voltage sequence data into the CNN model to obtain the SOH estimation result. Then, input the SOH estimation results output by the two models and the 5 relaxation voltage features into the LR model to obtain the final SOH estimation result. Step S2 specifically includes the following steps: S2-1: Based on the original dataset D1 obtained in step S1, calculate the maximum value V of the relaxation voltage sequence data in each cycle. max Minimum value V min Shannon entropy V entropy Average value V mean and standard deviation V std As features, they, along with the SOH label values, form the feature dataset D. f,1 And after normalization, its specific form is as follows: in, This represents the maximum relaxation voltage in the i-th cycle. This represents the minimum relaxation voltage in the i-th cycle. The Shannon entropy value of the relaxation voltage in the i-th cycle is represented by the following formula: , This represents the probability of the nth element appearing in a data sequence, where N represents the total number of values ​​in the data sequence. This represents the average relaxation voltage during the i-th cycle. This represents the standard deviation of the relaxation voltage in the i-th cycle; S2-2: Calculate the first-order differential voltage sequence data ∆V of the relaxed voltage sequence data. The specific form of the sequence data in the i-th cycle is as follows: in, The relaxation voltage sequence data V, the first-order differential voltage sequence data ∆V, and the corresponding SOH label values ​​are combined to form a feature dataset D. f,2 And after normalization, its specific form is as follows: 。 2. The SOH estimation method for lithium-ion batteries in an energy storage power station according to claim 1, characterized in that, Step S1 specifically includes the following steps: S1-1: Collect the relaxation voltage sequence data V of the lithium-ion battery during the 5-minute rest period after full charging in each cycle. The specific format of the sequence data for the i-th cycle is as follows: in, This represents the nth voltage value in the i-th cycle; S1-2: The relaxation voltage sequence data V and the SOH label value of the battery in each cycle are combined to form the original dataset D1, which is as follows: 。 3. The SOH estimation method for lithium-ion batteries in an energy storage power station according to claim 1, characterized in that, Step S4 specifically includes the following steps: S4-1: Build a LightGBM model and set the model's learning rate, number of estimators, and maximum tree depth; S4-2: Using the training set D obtained in step S3 train,1 and test set D test,1 The LightGBM model was trained and tested, and the mean absolute error (MAE) was used to measure the estimation accuracy of the LightGBM model. The formula for calculating MAE is as follows: in, This represents the SOH estimate of the battery in the l-th cycle, as output by the model. This represents the actual SOH value of the battery in the lth cycle; The LightGBM model completes training when its MAE on the test set is ≤2%; otherwise, repeat step S4 until this condition is met.

4. The SOH estimation method for lithium-ion batteries in an energy storage power station according to claim 1, characterized in that, Step S5 specifically includes the following steps: S5-1: Construct a one-dimensional convolutional neural network (CNN), and set the number of layers, number of convolutional kernels, kernel size, kernel stride, and learning rate. S5-2: Using the training set D obtained in step S3 train,2 and test set D test,2 The CNN model is trained and tested. The mean absolute error (MAE) is used to measure the estimation accuracy of the CNN model. When the MAE of the model on the test set is ≤2%, the model is considered to have completed training. Otherwise, step S5 is repeated until the condition is met.

5. The SOH estimation method for lithium-ion batteries in an energy storage power station according to claim 1, characterized in that, Step S6 specifically includes the following steps: S6-1: Extract the 5 relaxation voltage features from step S2, i.e., the maximum value V max Minimum value V min Shannon entropy V entropy Average value V mean and standard deviation V std And the LightGBM model on the feature dataset D f,1 The SOH estimation results output above, and the CNN model on the feature dataset D f,2 The SOH estimation results output above are merged and combined with the SOH label values ​​to form the feature dataset D. f,3 Its specific form is as follows: in, This represents the SOH estimate output by the LightGBM model in the i-th loop. This represents the SOH estimate output by the CNN model in the i-th loop; S6-2: Transfer the feature dataset D f,3 The dataset is partitioned according to the partitioning method in step S3 to obtain the training set D. train,3 and test set D test,3 .

6. The SOH estimation method for lithium-ion batteries in an energy storage power station according to claim 1, characterized in that, Step S7 specifically includes the following steps: S7-1: Construct a linear regression (LR) model; S7-2: Using the training set D obtained in step S6 train,3 and test set D test,3 The LR model is trained and tested. The mean absolute error (MAE) is used to measure the estimation accuracy of the LR model. When the MAE of the LR model on the test set is ≤1.5%, the model is considered to have completed training. Otherwise, step S7-2 is repeated until the condition is met.

Images