Feature screening method, system and device applied to lithium battery soh and rul prediction
By automatically filtering lithium battery charge-discharge cycle data features using the transformer model, the problem of relying on manual experience to select features in existing technologies is solved, and efficient and accurate prediction of lithium battery health and remaining service life is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TERMINUSBEIJING TECH CO LTD
- Filing Date
- 2023-03-23
- Publication Date
- 2026-05-12
AI Technical Summary
Existing methods for predicting the health and remaining lifespan of lithium batteries rely on manual experience to select features, which means that feature selection may not be optimal and features cannot be extracted before a fixed number of cycles is reached, resulting in a huge workload.
The Transformer model is used to learn lithium battery charge and discharge cycle data. The optimal features are automatically selected through the first mapping module, attention module and second mapping module in the Transformer module. The model parameters are optimized by using a deep learning regressor to achieve automatic feature selection.
It automatically selects features with better performance, reduces manual workload, solves the problem of not being able to extract features before reaching a fixed number of iterations, and improves prediction accuracy and efficiency.
Smart Images
Figure CN116577661B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of lithium battery health and remaining life prediction technology, and in particular to a feature screening method, system and device for predicting lithium battery health and remaining life. Background Technology
[0002] Lithium-ion battery health (SOH) is one of the indicators for measuring the current health status of a lithium-ion battery. It is generally measured by the ratio of the current capacity (Co) to the original capacity (Cn) at the time of manufacture. Lithium-ion battery remaining useful life (RUL) is an indicator for measuring the remaining useful life of a lithium-ion battery. When the SOH drops to a certain threshold, such as 80%, the lithium-ion battery is considered to have reached its end of life (EOL). The general unit for EOL is the number of cycles. RUL is EOL minus the number of cycles already used. Predicting lithium-ion battery health (SOH) and remaining useful life (RUL) is based on current / voltage or temperature data from the previous N cycles to predict the subsequent SOH and RUL of the lithium-ion battery.
[0003] Current mainstream technologies in the industry are mostly based on human experience, manually calculating features from raw data (such as current / voltage or temperature curves from the first N cycles), and then using these calculated features to train machine learning models, such as linear regression models, to predict the state of equilibrium (SOH) and relative uptime (RUL) of lithium batteries. [MIT-Standford, Nature Energy, 2019]
[0004] The most commonly used feature in existing prediction methods is the standard deviation of all points on the QV (capacity-voltage) difference curve. The difference comes from two different cycles, typically the difference between the 100th and 10th cycles. However, the numbers for the 100th and 10th cycles on the QV difference curve, and the standard deviation itself, are all chosen empirically. This selection has several problems:
[0005] (1) The selected feature calculation method is based on human experience and may not be the optimal solution. For example, it may not be the difference between the curves of the 100th and 10th orders, nor may it be the standard deviation statistic.
[0006] (2) The model cannot predict SOH and RUL when the battery has not been cycled 100 times.
[0007] (3) If the optimal solution of hyperparameters is to be obtained according to the current settings, it is necessary to traverse all sets for evaluation, which is a huge workload. Summary of the Invention
[0008] The purpose of this invention is to propose a feature screening method, system, and device for predicting the health and remaining lifespan of lithium batteries. This method can replace manual experience screening, quickly screen out the optimal features, and solve the problem that features cannot be extracted before a certain number of cycles are reached.
[0009] To achieve this objective, the present invention adopts the following technical solution:
[0010] A feature selection method for predicting the health and remaining lifespan of lithium batteries, the method comprising:
[0011] S1: Collect time-series cyclic data from the first to Nth charge-discharge cycles of the lithium battery, and divide the cyclic data into a training set and a test set;
[0012] S2: Build a transformer model, input the training set into the transformer model for training, and learn the optimal cyclic relationship and feature quantity;
[0013] S3: Input the test set into the trained transformer model and extract features.
[0014] As a preferred embodiment of the present invention, the transformer model built in S2 includes a single or M stacked transformer modules, wherein the transformer module includes a plurality of first mapping modules, an attention module and a plurality of second mapping modules;
[0015] The first mapping module is used to perform nonlinear transformation on the input cyclic data and learned feature data to generate feature vectors, and transmit them to the attention module;
[0016] The attention module is used to calculate the relative positions of all feature vectors output by the first mapping module, accumulate them, find the optimal cyclic relationship, and transmit it to the second mapping module.
[0017] The second mapping module is used to remap the cyclic relationship output by the attention module to find the optimal feature quantity.
[0018] As a preferred embodiment of the present invention, step S2, which involves inputting the training set into the transformer model for training, includes:
[0019] The training set and learned features are input into the first mapping module;
[0020] The last output of the second mapping module is taken as the output of the transformer model, and the output is input into the deep learning regressor to obtain the predicted health and remaining lifespan.
[0021] The predicted health and remaining lifespan are compared with the actual health and remaining lifespan using a loss function, and all parameters in the transformer model are optimized using a stochastic descent method.
[0022] As a preferred embodiment of the present invention, the loss function is mean squared error.
[0023] As a preferred embodiment of the present invention, the time-series-based cyclic data collected in S1 includes capacity, internal resistance, power or self-discharge resistance that affect the prediction of lithium battery health, as well as current and voltage that affect the prediction of the remaining lifespan of the lithium battery.
[0024] As a preferred embodiment of the present invention, the first mapping module is a fully connected layer, a direct mapping layer, or a convolutional layer, and the parameters of each first mapping module are shared.
[0025] As a preferred embodiment of the present invention, the learning feature data are parameters that the transformer module can learn, and their dimensions are the same as those of the lithium battery single-cycle data.
[0026] As a preferred embodiment of the present invention, the number of the first mapping module and the second mapping module is the same, which is equal to the sum of the number of lithium battery charge-discharge cycles and the number of learned features.
[0027] A feature screening system for predicting the health and remaining lifespan of lithium batteries, the system comprising:
[0028] The data acquisition module is used to collect time-series cyclic data of the lithium battery from the first to the Nth charge-discharge cycles, and divide the cyclic data into a training set and a test set.
[0029] The training module is used to input the training set into the transformer model for training, thereby learning the optimal cyclic relationships and feature quantities;
[0030] The feature extraction module is used to input the test set into the trained transformer model and extract features.
[0031] An apparatus includes a processor and a memory for storing processor-executable instructions; the processor implements the aforementioned feature screening method for predicting the health and remaining lifespan of lithium batteries by executing the executable instructions.
[0032] A computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the above-described feature screening method for predicting the health and remaining lifespan of lithium batteries.
[0033] The beneficial effects of this invention are as follows:
[0034] 1. The feature screening method, system and equipment of the present invention for predicting the health and remaining service life of lithium batteries can put the cyclic relationship (such as 10 times and 100 times) and feature quantities (such as standard deviation, variance, ratio, mean, etc.) of lithium battery charge and discharge cycle data into the transformer model for learning, which replaces manual experience screening, obtains features with better performance and saves the huge workload of manual feature screening.
[0035] 2. The feature screening method, system, and device of the present invention for predicting the health and remaining service life of lithium batteries are simple to train with a single transformer module. Multiple transformer modules can be trained to meet different lithium battery charge and discharge cycle requirements, which solves the problem that existing methods cannot extract features if a certain fixed number of cycles has not been reached. Attached Figure Description
[0036] Figure 1 This is one of the flowcharts of the feature screening method for predicting the health and remaining lifespan of lithium batteries provided by the present invention.
[0037] Figure 2 This is the second flowchart of the feature screening method for predicting the health and remaining lifespan of lithium batteries provided by this invention.
[0038] Figure 3 This is a schematic diagram of the feature screening system for predicting the health and remaining lifespan of lithium batteries provided by the present invention.
[0039] Figure 4 This is a schematic diagram of the structure of the device provided by the present invention. Detailed Implementation
[0040] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.
[0041] like Figure 1 As shown, this invention provides a feature selection method for predicting the residual health (ROH) and remaining useful life (RUL) of lithium batteries. The main implementing body is a feature selection system for predicting the residual health (ROH) and remaining useful life (RUL) of lithium batteries. The method includes the following steps:
[0042] S1: Collect time-series cyclic data from the first to Nth charge-discharge cycles of the lithium battery, and divide the cyclic data into a training set and a test set;
[0043] S2: Build a transformer model, input the training set into the transformer model for training, and learn the optimal cyclic relationship and feature quantity;
[0044] S3: Input the test set into the trained transformer model and extract features.
[0045] Specifically, in S1 above, the time-series-based cyclic data collected by the lithium battery sensor includes factors affecting the prediction of lithium battery health such as capacity, internal resistance, power, or self-discharge resistance, as well as factors affecting the prediction of the remaining lifespan of the lithium battery such as current and voltage. In this embodiment, the cyclic data is input into the feature selection model in the form of a one-dimensional time series.
[0046] The transformer model built in S2 above includes a single transformer module or M stacked transformer modules, such as... Figure 2 As shown, the transformer module includes several first mapping modules, an attention module, and several second mapping modules. The first mapping modules perform nonlinear transformations on the input cyclic data and learned feature data to generate feature vectors, which are then transmitted to the attention modules. The attention modules calculate the relative positions of all feature vectors output by the first mapping modules, accumulate them, find the optimal cyclic relationship, and transmit it to the second mapping modules. The second mapping modules remap the cyclic relationships output by the attention modules to find the optimal feature values.
[0047] The number of first and second mapping modules is the same, equal to the sum of the number of lithium battery charge / discharge cycles and the number of learned features. The parameters of each first mapping module are shared; each first mapping module can be a fully connected layer, a direct mapping layer, or a convolutional layer, etc. The learned features are parameters that the transformer module can learn; there can be one or more of them, and their dimension is the same as the dimension of the lithium battery's single-cycle data. The second mapping module is a fully connected layer. The internal parameters of the transformer module, such as the number of first and second mapping modules and the number of learned features, can be changed.
[0048] In S2 above, inputting the training set into the transformer model for training includes:
[0049] Input the training set and learned features into the first mapping module;
[0050] The last output of the second mapping module is taken as the output of the transformer model. This output is then input into the deep learning regressor to obtain the predicted health status (SOH) and remaining useful life (RUL).
[0051] The predicted state of health (SOH) and remaining useful life (RUL) are compared with the actual state of health (SOH) and remaining useful life (RUL) as a loss function, and all parameters in the transformer model are optimized using a stochastic descent approach. The loss function is the mean squared error.
[0052] The deep learning regressor is a fully connected layer or another commonly used deep learning layer. The loss function is preferably mean squared error.
[0053] In practice, the attention module within the transformer module captures the relationships between data from iterations 1 to N, learning the optimal cyclic relationships and thus avoiding the manual setting of empirical features for iterations 100 and 10. The second mapping module within the transformer module remaps the relationships obtained in the attention module, learning the optimal feature quantity, which can be standard deviation, variance, ratio, or mean, replacing the manually selected standard deviation as a feature. In other words, the transformer module uses deep learning to optimize the health (SOH) and remaining useful life (RUL) objectives, enabling the learning quantities in the attention module and the second mapping module to learn optimal solutions.
[0054] Training a single transformer module is easy, and different transformer modules can be trained for different number of cycles N, making feature combinations more flexible. For example, when a lithium battery has only been used for 3 cycles, a transformer module with N=3 can be used to extract features. The extracted features are then input into a deep learning regressor to predict the health status (SOH) and remaining lifespan (RUL), thus solving the problem that existing methods cannot extract features before reaching a certain fixed number of cycles (such as 100 cycles).
[0055] In addition, by incorporating manually acquired features such as the number of charge-discharge cycles associated with lithium batteries (e.g., 10 versus 100 cycles) and characteristic quantities (e.g., standard deviation, variance, ratio, or mean) into the transformer module for learning, the feature selection process replaces manual experience-based selection, resulting in better-performing features and saving the enormous workload of manual feature selection.
[0056] like Figure 3As shown, the present invention also provides a feature selection system for predicting the state of health (SOH) and remaining useful life (RUL) of lithium batteries. This system includes a data acquisition module 110, a training module 120, and a feature extraction module 130. The data acquisition module 110 collects time-series cyclic data from the first to Nth charge-discharge cycles of the lithium battery, dividing this cyclic data into a training set and a test set. The training module 120 inputs the training set into a transformer model for training, thereby learning the optimal cyclic relationships and feature quantities. The feature extraction module 130 inputs the test set into the trained transformer model to extract features.
[0057] The transformer model comprises a single or M stacked transformer modules. Each transformer module includes a first mapping module, an attention module, and a second mapping module. The first mapping module performs non-linear transformations on the input recurrent data and learned feature data to generate feature vectors, which are then passed to the attention module. The attention module calculates the relative positions of all feature vectors output by the first mapping module, accumulates them, finds the optimal recurrent relationship, and passes it to the second mapping module. The second mapping module remaps the recurrent relationship output by the attention module to find the optimal feature quantity.
[0058] It should be noted that the system provided in this embodiment can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.
[0059] like Figure 4 As shown, the present invention also provides a device including a processor 210, a communication interface 220, a memory 230 for storing processor-executable instructions, and a communication bus 240. The processor 210, communication interface 220, and memory 230 communicate with each other via the communication bus 240. The processor 210 executes executable instructions to implement the aforementioned feature screening method for predicting lithium battery health (SOH) and remaining useful life (RUL).
[0060] The logical instructions in memory 230 can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0061] The present invention also provides a computer-readable storage medium storing computer instructions thereon, which, when executed by a processor, implement the above-described feature selection method for predicting lithium battery health (SOH) and remaining useful life (RUL).
[0062] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected based on actual needs to achieve the purpose of this embodiment. Those skilled in the art can understand and implement this without any creative effort.
[0063] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.
[0064] The technical principles of the present invention have been described above with reference to specific embodiments. These descriptions are merely for explaining the principles of the invention and should not be construed as limiting the scope of protection of the invention in any way. Based on this explanation, those skilled in the art can readily conceive of other specific embodiments of the invention without inventive effort, and these embodiments will all fall within the scope of protection of the present invention.
Claims
1. A feature selection method for predicting the health and remaining lifespan of lithium batteries, characterized in that, The method includes: S1: Collect time-series cyclic data from the first to Nth charge-discharge cycles of the lithium battery, and divide the cyclic data into a training set and a test set; S2: Build a transformer model, input the training set into the transformer model for training, thereby learning the optimal cyclic relationship and feature quantity; wherein, the feature quantity includes at least one of standard deviation, variance, ratio and mean; S3: Input the test set into the trained transformer model and extract features.
2. The feature screening method for predicting the health and remaining lifespan of lithium batteries according to claim 1, characterized in that, The transformer model built in S2 includes a single or M stacked transformer modules, and each transformer module includes several first mapping modules, attention modules, and several second mapping modules. The first mapping module is used to perform nonlinear transformation on the input cyclic data and learned feature data to generate feature vectors, and transmit them to the attention module; The attention module is used to calculate the relative positions of all feature vectors output by the first mapping module, accumulate them, find the optimal cyclic relationship, and transmit it to the second mapping module. The second mapping module is used to remap the cyclic relationship output by the attention module to find the optimal feature quantity.
3. The feature screening method for predicting the health and remaining lifespan of lithium batteries according to claim 2, characterized in that, In step S2, inputting the training set into the transformer model for training includes: The training set and learned features are input into the first mapping module; The last output of the second mapping module is taken as the output of the transformer model, and the output is input into the deep learning regressor to obtain the predicted health and remaining lifespan. The predicted health and remaining lifespan are compared with the actual health and remaining lifespan using a loss function, and all parameters in the transformer model are optimized using a stochastic descent method.
4. The feature screening method for predicting the health and remaining lifespan of lithium batteries according to claim 3, characterized in that, The loss function is the mean squared error.
5. The feature screening method for predicting the health and remaining lifespan of lithium batteries according to claim 1, characterized in that, The time-series-based cyclic data collected in S1 includes capacity, internal resistance, power, or self-discharge resistance that affect the prediction of lithium battery health, as well as current and voltage that affect the prediction of the remaining lifespan of the lithium battery.
6. The feature screening method for predicting the health and remaining lifespan of lithium batteries according to claim 2, characterized in that, The first mapping module is a fully connected layer, a direct mapping layer, or a convolutional layer, and the parameters of each first mapping module are shared.
7. The feature screening method for predicting the health and remaining lifespan of lithium batteries according to claim 2, characterized in that, The learned feature data are parameters that the transformer module can learn, and their dimensions are the same as those of the lithium battery single-cycle data.
8. The feature screening method for predicting the health and remaining service life of lithium batteries according to claim 2, characterized in that, The number of the first mapping module and the second mapping module is the same, equal to the sum of the number of lithium battery charge-discharge cycles and the number of learned features.
9. A feature screening system for predicting the health and remaining lifespan of lithium batteries, characterized in that, The system includes: The data acquisition module is used to collect time-series cyclic data of the lithium battery from the first to the Nth charge-discharge cycles, and divide the cyclic data into a training set and a test set. The training module is used to input the training set into the transformer model for training, thereby learning the optimal cyclic relationship and feature quantities; wherein, the feature quantities include at least one of standard deviation, variance, ratio and mean; The feature extraction module is used to input the test set into the trained transformer model and extract features.
10. An apparatus comprising a processor and a memory for storing processor-executable instructions; characterized in that, The processor executes the executable instructions to implement the feature screening method for predicting the health and remaining lifespan of lithium batteries as described in any one of claims 1-8.