Lithium ion battery sensorless temperature estimation method and system

Abstract

The invention discloses a sensorless temperature estimation method and system for a lithium ion battery. The method comprises the following steps: dividing an existing data set into a source domain data set and a target domain data set; performing data processing on voltage and current data in the source domain data set; inputting the temperature signal in the source domain data set and the voltage and current data after data enhancement into an iTransform model for pre-training, and storing; performing a data processing process on voltage and current data in the target domain data set; constructing a model similar to the model in the previous step, and copying parameters of the pre-trained model; and inputting the processed voltage and current data and the first 30% of temperature data in the target domain data set into the new model for retraining, and finally predicting the last 70% of temperature curve of the target domain data set. The system comprises a data diversity module, a source domain data processing module, a target domain data processing module, a new model construction module and a prediction module. According to the method, the precision of temperature estimation can be ensured, and the generalization ability of the model can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium-ion batteries, and in particular relates to a sensorless temperature estimation method and system for lithium-ion batteries. Background Art

[0002] Under normal circumstances, the DC power supply system of a substation provides power for important loads such as control and protection in the substation. In the event of a substation failure and the loss of AC power supply to the station, it continuously and reliably provides power to the DC load within a specified time, achieves rapid fault removal and recovery, and ensures the safety of personnel and equipment. Therefore, a reliable DC power supply system is the key to ensuring the normal operation of the substation, and its safe and stable operation is directly related to the reliable power supply of the main equipment of the substation. In the DC power supply system, batteries play a core role. Lithium-ion batteries have the advantages of long life, good temperature characteristics, and high reliability, and are expected to be an ideal replacement for lead-acid batteries in DC power supply systems.

[0003] The safety issues of lithium-ion batteries cannot be ignored, especially the temperature state of the battery. Battery performance deteriorates at high temperatures, and the rate of aging and capacity decay will accelerate; in low temperature environments, the chemical reaction of the battery will also slow down significantly, and the discharge efficiency will decrease. This reflects the necessity of monitoring the temperature of the battery. In high-power applications, temperature sensors are usually used to monitor the temperature. The battery management system adjusts the working state of the battery according to the readings of the temperature sensor to prevent the temperature from being too high or too low. However, in large-scale battery packs, if a temperature sensor is installed for each battery, the cost will increase significantly, and it will take up additional space, affecting the overall layout and heat dissipation design of the system. Therefore, many battery systems choose to install a limited number of temperature sensors at key locations in the battery pack instead of covering all batteries. This may cause local overheating of some cells or battery modules to not be detected in time, causing safety issues. Therefore, obtaining the approximate temperature of the battery through an effective sensorless temperature estimation method can be used to reduce the number of sensors required for the battery pack, which helps to reduce costs and improve reliability.

[0004] There are three main methods for sensorless temperature estimation: estimation based on electrochemical impedance spectroscopy, estimation based on battery thermal model, and estimation based on data-driven. Electrochemical impedance spectroscopy is to determine the impedance of the battery at a specific frequency by applying a small AC voltage or current signal and measuring the response of the system, so as to infer the temperature of the battery. However, the hardware required for electrochemical impedance spectroscopy measurement is complex and costly, and requires complex processing and modeling. Whether the impedance can be obtained in real time is also a huge challenge. Estimation based on thermal model establishes a thermal-electric coupling model of the battery, combines the current, voltage and other electrical characteristics of the battery, and infers the temperature of the battery. However, this requires accurate models and parameters (such as thermal conductivity, heat capacity, etc.), and model errors will affect the estimation accuracy, making it very difficult to implement in practice. Estimation based on data-driven training establishes the mapping relationship between the electrical parameters of the battery (such as current, voltage, internal resistance, etc.) and temperature by training the machine learning model. It trains the model with a large amount of data and can adapt to complex battery characteristics and nonlinear relationships. However, the existing methods use too many battery data variables, which makes the model complex and increases the difficulty of training. Reducing variables will bring about the problems of low model accuracy and insufficient model generalization ability. Summary of the invention

[0005] The purpose of the present invention is to provide a sensorless temperature estimation method and system for lithium-ion batteries. Data enhancement improves the generalization ability and robustness of the model by generating more training data, and transfer learning uses existing knowledge and pre-trained models to greatly improve the training efficiency, performance and generalization ability of the model. Combining data enhancement with transfer learning can not only ensure the accuracy of temperature estimation, but also improve the generalization ability of the model.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A sensorless temperature estimation method for a lithium-ion battery, comprising:

[0008] The first step is to divide the existing dataset into source domain dataset and target domain dataset;

[0009] The second step is to process the voltage and current data in the source domain data set. The data processing includes SG and mean filtering, extracting data differential features, and enhancing the data.

[0010] The third step is to input the temperature signal in the source domain data set and the voltage and current data after data enhancement into the iTransform model for pre-training, and save the pre-trained model;

[0011] The fourth step is to process the voltage and current data in the target domain data set, that is, to perform SG and mean filtering on the voltage and current signals, extract data differential features, and enhance the data;

[0012] Step 5: Build a model similar to that in step 3 and copy the parameters of the pre-trained model in step 3.

[0013] In the sixth step, the voltage and current data processed in the fourth step and the temperature data of the first 30% of the target domain data set are input into the new model in the fifth step for retraining, the number of training rounds and the learning rate of the model are reduced to fine-tune the model parameters, and finally the temperature curve of the last 70% of the target domain data set is predicted.

[0014] A further improvement of the present invention is that, in the first step, the source domain data set is used to pre-train model parameters, and the target domain data set is used to predict the temperature after the model is trained.

[0015] A further improvement of the present invention is that in the second step, data enhancement includes two steps: adding random noise to the data and quantizing the signal.

[0016] A further improvement of the present invention is that, in the second step, SG and mean filtering are performed on the voltage and current data, and the formula for SG filtering is:

[0017] y=f u (x)(1)

[0018] where f u represents the filtering operation, i.e., the mapping of input x values ​​to output y values, and μ is the SG filter parameter, including the fitting order and window size;

[0019] The formula for mean filtering is:

[0020] y=mean(S)(2)

[0021] Where S is the data in the mean filter window, and mean is the mean algorithm.

[0022] A further improvement of the present invention is that, in the third step, for the iTransform model, given a multidimensional time series with a time series length of T and a number of variables of N Using X :,n Represents the entire sequence of the same variable; use the embedding layer to learn the variable X :,n The sequence feature representation of , independently aggregates the global features of each variable:

[0023]

[0024] in It contains all the time series changes of the corresponding variable in the past time, which is called the variable marking part; in the subsequent layers, each variable marking part exchanges information through the self-attention mechanism; within each variable marking part, the layer normalization unifies the measurement units and feature distributions of different variables, and the feedforward network performs feature encoding in a fully connected manner; finally, each variable marking part is mapped to the prediction result through the mapping layer; the whole calculation result is expressed as follows:

[0025] H l+1 =TranBlock(H l ),l=0,...,L-1(11)

[0026]

[0027] Where L is the number of layers in the Transformer model, the embedding layer: and the mapping layer: All of them are realized through multi-layer perceptron. is the prediction result.

[0028] A sensorless temperature estimation system for a lithium-ion battery, comprising:

[0029] The data segmentation module divides the existing data set into a source domain data set and a target domain data set;

[0030] The first source domain data processing module processes the voltage and current data in the source domain data set. The data processing includes SG and mean filtering, extracting data differential features, and enhancing the data.

[0031] The second source domain data processing module inputs the temperature signal in the source domain data set and the voltage and current data after data enhancement into the iTransform model for pre-training, and saves the pre-trained model;

[0032] The target domain data processing module processes the voltage and current data in the target domain data set, that is, performs SG and mean filtering on the voltage and current signals, extracts data differential features, and enhances the data;

[0033] A new model building module is used to build a model of the same type as that in the second source domain data processing module, and to copy the parameters of the pre-trained model in the second source domain data processing module;

[0034] The prediction module inputs the voltage and current data processed in the target domain data processing module and the temperature data of the first 30% of the target domain data set into the new model in the new model construction module for retraining, reduces the number of model training rounds and the learning rate to fine-tune the model parameters, and finally predicts the temperature curve of the last 70% of the target domain data set.

[0035] A further improvement of the present invention is that, in the data diversity module, the source domain data set is used to pre-train model parameters, and the target domain data set is used to predict the temperature after the model is trained.

[0036] A further improvement of the present invention is that, in the first source domain data processing module, data enhancement includes two steps: adding random noise to the data and quantizing the signal.

[0037] A further improvement of the present invention is that, in the first source domain data processing module, SG and mean filtering are performed on the voltage and current data, and the formula for SG filtering is:

[0038] y=f u (x)(1)

[0039] where f u represents the filtering operation, i.e., the mapping of input x values ​​to output y values, and μ is the SG filter parameter, including the fitting order and window size;

[0040] The formula for mean filtering is:

[0041] y=mean(S)(2)

[0042] Where S is the data in the mean filter window, and mean is the mean algorithm.

[0043] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the sensorless temperature estimation method for a lithium-ion battery are implemented.

[0044] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0045] The first point is the data preprocessing method. Use SG filtering to enhance the stability of the data, and use mean filtering to increase the fluctuation range of the curve, so that it is closer to the fluctuation of the real battery data curve. Extract the differential features of the data and explore the relationship between temperature and voltage and current. Perform data enhancement (add random noise and quantization) on the data to improve the generalization ability and robustness of the model. Among them, SG filtering is used to smooth the data through local polynomial regression, effectively removing high-frequency noise, and mean filtering is used to reduce data fluctuations by calculating the average value of local data points. Improve the quality and reliability of the data, solve the problem that the data cannot fit the actual battery situation well, and enable the model to more accurately capture the trend of battery performance changes. By extracting differential features and adding random noise to the data and quantizing it, the quality problems of the data, such as insufficient data volume, uneven data distribution, low data quality, high model complexity, or large differences between training data and actual application scenarios, are solved. The model performs well on training data, but performs poorly on new and unseen data, so that the model achieves the expected generalization ability and robustness.

[0046] The second point is the use of transfer learning and the adoption of the iTransformer transferable temperature estimation model, which, combined with data processing methods, can effectively estimate the temperature using only voltage and current data. Due to the use of transfer learning and the adoption of the inverted transformer transferable temperature estimation model, the accuracy problem of temperature estimation using only voltage and current data is solved, and the expected effect of effectively estimating battery temperature without relying on additional temperature sensors is achieved. This model, combined with data processing methods, can provide more accurate temperature predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A flow chart of a sensorless temperature estimation method for a lithium-ion battery according to the present invention;

[0049] Figure 2 It is a schematic diagram of the SG filtering and mean filtering results of the current data;

[0050] Figure 3 is the differential characteristic curve of current data;

[0051] Figure 4 This is a schematic diagram of current data after adding noise;

[0052] Figure 5 It is a schematic diagram of quantified current data;

[0053] Figure 6 This is a schematic diagram of temperature prediction results;

[0054] Figure 7 The present invention is a structural block diagram of a lithium-ion battery sensorless temperature estimation system. DETAILED DESCRIPTION

[0055] In the following, only some exemplary embodiments are briefly described. As those skilled in the art will appreciate, the described embodiments may be modified in various ways without departing from the spirit or scope of the present invention. Therefore, the drawings and descriptions are considered to be exemplary and non-restrictive in nature.

[0056] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0057] It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0058] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0059] Various structural schematic diagrams of the embodiments disclosed in the present invention are shown in the accompanying drawings. These figures are not drawn to scale, and some details are magnified and some details may be omitted for the purpose of clear expression. The shapes of various regions and layers shown in the figures and the relative sizes and positional relationships therebetween are only exemplary, and may deviate in practice due to manufacturing tolerances or technical limitations, and those skilled in the art may additionally design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0060] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0061] Example 1

[0062] like Figure 1As shown, the present invention provides a sensorless temperature estimation method for lithium-ion batteries, in which the data set includes the voltage, current, and temperature signals of the battery. The present invention will predict the temperature based on the voltage and current signals. Specifically, the following steps are included:

[0063] The first step is to divide the existing dataset into a source domain dataset and a target domain dataset. The source domain dataset is used to pre-train the model parameters, and the target domain dataset is used to predict the temperature after the model is trained.

[0064] The second step is to process the voltage and current data in the source domain data set. Data processing includes SG and mean filtering, extracting data differential features, and enhancing the data. Data enhancement is to improve the generalization ability and robustness of the model. Data enhancement includes two steps: adding random noise to the data and quantizing the signal.

[0065] In the third step, the temperature signal in the source domain data set and the voltage and current data after data enhancement are input into the iTransform model for pre-training, and the pre-trained model is saved.

[0066] In the fourth step, the voltage and current data in the target domain data set are subjected to the same data processing process as in the second step, that is, SG and mean filtering are performed on the voltage and current signals, data differential features are extracted, and data enhancement is performed.

[0067] In the fifth step, we build a model similar to that in the third step and copy the parameters of the pre-trained model in the third step.

[0068] In the sixth step, the voltage and current data processed in the fourth step and the temperature data of the first 30% of the target domain data set are input into the new model in the fifth step for retraining, the number of training rounds and the learning rate of the model are reduced to fine-tune the model parameters, and finally the temperature curve of the last 70% of the target domain data set is predicted.

[0069] Example 2

[0070] In this embodiment, after processing the voltage and current data, a model is trained to estimate the temperature.

[0071] 1. Perform SG and mean filtering on the voltage and current data. The formula for SG filtering is:

[0072] y=f u (x)(1)

[0073] where f u represents the filtering operation, that is, the mapping of input x values ​​to output y values, and μ is the SG filter parameter, including the fitting order and window size.

[0074] The formula for mean filtering is:

[0075] y=mean(S)(2)

[0076] Where S is the data in the mean filter window, and mean is the mean algorithm.

[0077] 2. Use the differential method to calculate the difference between adjacent values ​​of the voltage and current curve after SG filtering, and extract the time series data that is correlated with temperature. The algorithm formula is:

[0078]

[0079] in is the current time y t With the previous moment y t-1 The difference value of the current data after filtering and differential feature extraction is shown in the figure below. Figure 2 and Figure 3 shown.

[0080] 3. The random noise generated is x t (t=1,2,3…n), n is the number of samples in the data set. The process of generating random noise requires two parameters: mean and standard deviation. The mean formula is as follows:

[0081]

[0082] in is the mean of the generated random noise data, and the standard deviation formula is as follows:

[0083]

[0084] Where σ is the standard deviation of the generated random noise. t (t=1,2,3…n) add to obtain enhanced sequence data z t (t=1,2,3…n), the formula is as follows:

[0085] z t =x t +y t ,(t=1,2,3...n)(6)

[0086] 4. Quantize the time series to a level set, input the number of levels w to be divided, and calculate the maximum value max and minimum value min in the data set respectively. The formula is as follows.

[0087]

[0088] Where d is the difference in level, and the level set A = {a1, a2, …, aw}. The numbers in the set are arranged from small to large, where a1 = min + d / 2, and the quantization formula is as follows:

[0089]

[0090] The current data after adding noise and quantization are as follows Figure 4 and Figure 5 shown.

[0091] 5. After the voltage and current data in the source domain data are preprocessed in steps 1-4, they are used together with the temperature data in the source domain data as input to pre-train the iTransformer model and save the training parameters. For the iTransformer model, given a multidimensional time series with a time series length of T and a number of variables of N Using X :,n Represents the entire sequence of the same variable. Use the embedding layer to learn the variable X :,n The sequence feature representation of , independently aggregates the global features of each variable:

[0092]

[0093] in It contains all the time series changes of the corresponding variable in the past time, which is called the variable marking part. In the subsequent layers, each variable marking part exchanges information through the self-attention mechanism. Within each variable marking part, the layer normalization unifies the measurement units and feature distributions of different variables, and the feedforward network performs feature encoding in a fully connected manner. Finally, each variable marking part is mapped to the prediction result through the mapping layer. The whole calculation result is expressed as follows:

[0094] H l+1 =TranBlock(H l ),l=0,...,L-1(11)

[0095]

[0096] Where L is the number of layers in the Transformer model, the embedding layer: and the mapping layer: All of them are implemented through multi-layer perceptron (MLP). is the prediction result.

[0097] 6. Copy a similar model in 5 and copy the parameters of the pre-trained model in 5. After the current and voltage data in the target domain data are processed by steps 1-4, they are combined with the temperature data of the first 30% of the target domain data set and input into this model for retraining. The number of training rounds and the learning rate of the model are reduced to fine-tune the model parameters. Finally, the temperature curve of the last 70% of the target domain data set is predicted. The temperature prediction results of the method proposed in the present invention are as follows: Figure 6 As shown in the figure, the temperature prediction results of the recurrent neural network, long short-term memory network and gated recurrent unit for the same data are compared. In order to highlight the accuracy and generalization ability of the proposed model, a comparative experiment of three models (RNN, LSTM and GRU) was conducted. The comparative experimental model adopts the same migration strategy as the proposed model. First, the three machine learning methods are used for pre-training on the source domain data, and then the pre-trained model is copied and retrained using the first 30% data of the target domain to predict the temperature of the last 70% of the target domain, and then compared with the prediction effect of the proposed method.

[0098] Example 3

[0099] like Figure 7 As shown, the present invention provides a lithium-ion battery sensorless temperature estimation system, comprising:

[0100] The data segmentation module divides the existing data set into a source domain data set and a target domain data set;

[0101] The first source domain data processing module processes the voltage and current data in the source domain data set. The data processing includes SG and mean filtering, extracting data differential features, and enhancing the data.

[0102] The second source domain data processing module inputs the temperature signal in the source domain data set and the voltage and current data after data enhancement into the iTransform model for pre-training, and saves the pre-trained model;

[0103] The target domain data processing module processes the voltage and current data in the target domain data set, that is, performs SG and mean filtering on the voltage and current signals, extracts data differential features, and enhances the data;

[0104] A new model building module is used to build a model of the same type as that in the second source domain data processing module, and to copy the parameters of the pre-trained model in the second source domain data processing module;

[0105] The prediction module inputs the voltage and current data processed in the target domain data processing module and the temperature data of the first 30% of the target domain data set into the new model in the new model construction module for retraining, reduces the number of model training rounds and the learning rate to fine-tune the model parameters, and finally predicts the temperature curve of the last 70% of the target domain data set.

[0106] Example 4

[0107] The present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the sensorless temperature estimation method for a lithium-ion battery are implemented.

[0108] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0109] The present application is described with reference to the flowcharts and / or block diagrams of the methods, systems and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A system that specifies the functions of a box or multiple boxes.

[0110] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0112] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved.

[0113] In addition, it should be understood that although this specification is described in accordance with the implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation modes that can be understood by those skilled in the art. The above content is only to illustrate the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution according to the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A sensorless temperature estimation method for a lithium-ion battery, characterized in that: include: The first step is to divide the existing dataset into source domain dataset and target domain dataset; The second step is to process the voltage and current data in the source domain data set. The data processing includes SG and mean filtering, extracting data differential features, and enhancing the data. The third step is to input the temperature signal in the source domain data set and the voltage and current data after data enhancement into the iTransform model for pre-training, and save the pre-trained model; The fourth step is to process the voltage and current data in the target domain data set, that is, to perform SG and mean filtering on the voltage and current signals, extract data differential features, and enhance the data; Step 5: Build a model similar to that in step 3 and copy the parameters of the pre-trained model in step 3. In the sixth step, the voltage and current data processed in the fourth step and the temperature data of the first 30% of the target domain data set are input into the new model in the fifth step for retraining, the number of training rounds and the learning rate of the model are reduced to fine-tune the model parameters, and finally the temperature curve of the last 70% of the target domain data set is predicted.

2. A sensorless temperature estimation method for lithium-ion batteries according to claim 1, characterized in that: In the first step, the source domain dataset is used to pre-train the model parameters, and the target domain dataset is used to predict the temperature after the model is trained.

3. A sensorless temperature estimation method for lithium-ion batteries according to claim 1, characterized in that: In the second step, data augmentation includes two steps: adding random noise to the data and quantizing the signal.

4. A sensorless temperature estimation method for lithium-ion batteries according to claim 1, characterized in that: In the second step, SG and mean filtering are performed on the voltage and current data. The formula for SG filtering is: y=f u (x)(1)where f u represents the filtering operation, i.e., the mapping of input x values ​​to output y values, and μ is the SG filter parameter, including the fitting order and window size; The formula for mean filtering is: y=mean(S)(2) where S is the data in the mean filter window and mean is the mean algorithm.

5. A sensorless temperature estimation method for lithium-ion batteries according to claim 1, characterized in that: In the third step, for the iTransform model, given a multidimensional time series with a time series length of T and a number of variables of N Using X :,n Represents the entire sequence of the same variable; use the embedding layer to learn the variable X :,n The sequence feature representation of , independently aggregates the global features of each variable: in It contains all the time series changes of the corresponding variable in the past time, which is called the variable marking part. In the subsequent layers, each variable marking part exchanges information through the self-attention mechanism. Within each variable tag part, the layer normalization unifies the measurement units and feature distribution of different variables, and the feedforward network performs feature encoding in a fully connected manner; finally, each variable tag part is mapped to a prediction result through the mapping layer; the entire calculation result is expressed as follows: H l+1 =TranBlock(H l ),l=0,…,L-1(11) Where L is the number of layers in the Transformer model, the embedding layer: and the mapping layer: All of them are realized through multi-layer perceptron. is the prediction result.

6. A sensorless temperature estimation system for a lithium-ion battery, characterized in that: include: The data segmentation module divides the existing data set into a source domain data set and a target domain data set; The first source domain data processing module processes the voltage and current data in the source domain data set. The data processing includes SG and mean filtering, extracting data differential features, and enhancing the data. The second source domain data processing module inputs the temperature signal in the source domain data set and the voltage and current data after data enhancement into the iTransform model for pre-training, and saves the pre-trained model; The target domain data processing module processes the voltage and current data in the target domain data set, that is, performs SG and mean filtering on the voltage and current signals, extracts data differential features, and enhances the data; A new model building module is used to build a model of the same type as that in the second source domain data processing module, and to copy the parameters of the pre-trained model in the second source domain data processing module; The prediction module inputs the voltage and current data processed in the target domain data processing module and the temperature data of the first 30% of the target domain data set into the new model in the new model construction module for retraining, reduces the number of model training rounds and the learning rate to fine-tune the model parameters, and finally predicts the temperature curve of the last 70% of the target domain data set.

7. A lithium-ion battery sensorless temperature estimation system according to claim 6, characterized in that: In the data diversity module, the source domain dataset is used to pre-train the model parameters, and the target domain dataset is used to predict the temperature after the model is trained.

8. A lithium-ion battery sensorless temperature estimation system according to claim 6, characterized in that: In the first source domain data processing module, data enhancement includes two steps: adding random noise to the data and quantizing the signal.

9. A lithium-ion battery sensorless temperature estimation system according to claim 6, characterized in that: In the first source domain data processing module, SG and mean filtering are performed on the voltage and current data. The formula for SG filtering is: y=f u (x)(1)where f u represents the filtering operation, i.e., the mapping of input x values ​​to output y values, and μ is the SG filter parameter, including the fitting order and window size; The formula for mean filtering is: y=mean(S)(2) where S is the data in the mean filter window and mean is the mean algorithm.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of a lithium-ion battery sensorless temperature estimation method according to any one of claims 1 to 5 are implemented.

Images