Battery capacity prediction method, model training method, electronic device and storage medium

Abstract

The embodiments of the present application provide a battery capacity prediction method, a model training method, an electronic device, and a storage medium. In the embodiments of the present application, on the one hand, the feature vectors of the manufacturing data representing the time series category of the battery are processed based on the anomaly detection model to detect whether the battery capacity is normal or abnormal. On the other hand, when the abnormal detection result output by the anomaly detection model indicates that the battery capacity is normal, the feature vectors of the manufacturing data representing the time series category of the battery and the feature vectors of the manufacturing data representing the discrete category of the battery are predicted and processed based on the regression model to predict the battery capacity. Thus, a data-driven battery capacity classification method is provided, which can predict the battery capacity of the battery with high precision by analyzing the manufacturing data of the lithium battery and combining the anomaly detection model and the regression model, thereby reducing the power consumption cost and time consumption, simplifying the entire manufacturing process of the battery, and improving the production efficiency and production safety of the battery.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a battery capacity prediction method, a model training method, an electronic device, and a storage medium. Background Art

[0002] Battery capacity refers to the amount of charge a battery can store and is a key performance metric for measuring battery performance. Most lithium battery manufacturers typically use the following method to scale lithium battery capacity: after formation, the battery is first charged to full capacity and then discharged to determine its capacity. However, this method requires a charging and discharging process, which is costly and time-consuming, and can pose a fire hazard, impacting lithium battery production efficiency and safety. Summary of the Invention

[0003] Multiple aspects of the present application provide a battery capacity prediction method, a model training method, an electronic device, and a storage medium for predicting the battery capacity of a battery with high precision, reducing power consumption costs and time consumption, and improving battery production efficiency and production safety.

[0004] An embodiment of the present application provides a battery capacity prediction method, comprising: obtaining manufacturing data of a time series category and manufacturing data of a discrete category of a target battery, the manufacturing data of the time series category representing data generated by the target battery during the manufacturing process, and the manufacturing data of the discrete category representing data of the manufacturing process of the target battery; performing feature engineering processing on the manufacturing data of the time series category and the manufacturing data of the discrete category, respectively, to extract a first eigenvector corresponding to the manufacturing data of the time series category and a second eigenvector corresponding to the manufacturing data of the discrete category; inputting the first eigenvector into an anomaly detection model to perform anomaly detection on the battery capacity of the target battery through the anomaly detection model; if the anomaly detection result indicates that the battery capacity of the target battery is normal, inputting the first eigenvector and the second eigenvector into a regression model to predict the battery capacity of the target battery through the regression model.

[0005] An embodiment of the present application also provides a model training method, including: obtaining sample time series feature vectors of multiple sample batteries; and performing model training on an initial anomaly detection model based on the sample feature vectors of the multiple sample batteries to obtain an anomaly detection model.

[0006] An embodiment of the present application also provides another model training method, including: obtaining sample time series feature vectors, sample discrete feature vectors and expected battery capacity of multiple sample batteries, the battery capacity of the sample batteries is normal, the sample time series feature vectors are obtained by feature engineering the manufacturing data of the time series category of the sample batteries, and the sample discrete feature vectors are obtained by feature engineering the manufacturing data of the discrete category of the sample batteries; using the sample time series feature vectors, sample discrete feature vectors and expected battery capacity of multiple sample batteries, an initial regression model is trained to obtain a regression model.

[0007] An embodiment of the present application also provides an electronic device, comprising: a memory and a processor; the memory is used to store a computer program; the processor is coupled to the memory and is used to execute the computer program to perform the steps in the battery capacity prediction method.

[0008] An embodiment of the present application further provides a computer-readable storage medium storing a computer program, including: when the computer program is executed by a processor, the processor is enabled to implement the steps in the battery capacity prediction method.

[0009] The technical solution provided by the embodiment of the present application, on the one hand, processes the feature vectors of the manufacturing data representing the time series category of the battery based on the anomaly detection model to detect whether the battery capacity is normal or abnormal. On the other hand, when the abnormal detection result output by the anomaly detection model indicates that the battery capacity is normal, the feature vectors of the manufacturing data representing the time series category of the battery and the feature vectors of the manufacturing data representing the discrete category of the battery are predicted and processed based on the regression model to predict the battery capacity. Thus, a data-driven battery capacity classification method is provided. By analyzing the manufacturing data of lithium batteries and combining the anomaly detection model and the regression model, the battery capacity of the battery can be predicted with high precision, reducing power consumption costs and time consumption, simplifying the entire battery manufacturing process, and improving battery production efficiency and production safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0011] Figure 1 An exemplary application scenario diagram provided for an embodiment of the present application;

[0012] Figure 2 A flowchart of a battery capacity prediction method provided in an embodiment of the present application;

[0013] Figure 3 A model structure diagram of an anomaly detection model provided in an embodiment of the present application;

[0014] Figure 4 A schematic diagram of a battery capacity prediction method provided in an embodiment of the present application;

[0015] Figure 5 Another schematic diagram of battery capacity prediction provided in an embodiment of the present application;

[0016] Figure 6 A schematic diagram of the structure of a battery capacity prediction device provided in an embodiment of the present application;

[0017] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0019] In the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the access relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. In the text description of the present application, the character " / " generally indicates that the previous and next associated objects are in an "or" relationship. In addition, in the embodiments of the present application, "first", "second", "third", etc. are only used to distinguish the contents of different objects and have no other special meanings.

[0020] The current capacity division method requires a charging and discharging process, which consumes a lot of electricity, is time-consuming, and is prone to fire hazards, affecting the production efficiency and safety of lithium batteries.

[0021] In addition, the industry generally believes that the electrochemical properties of lithium batteries are basically formed after the formation process and high-temperature static process. For example, a passivation layer, also known as the SEI film (solid electrolyte interphase), forms on the surface of the negative electrode during the formation process. The quality of the SEI film directly affects the battery's electrochemical properties such as cycle life, stability, self-discharge, and safety. The amount of electrolyte injection and the amount of electrolyte consumed during the formation process can also indicate the stability and capacity characteristics of the battery.

[0022] To this end, the embodiments of the present application provide a battery capacity prediction method, a model training method, an electronic device, and a storage medium. In the embodiments of the present application, on the one hand, the feature vectors of the manufacturing data representing the time series category of the battery are processed based on the anomaly detection model to detect whether the battery capacity is normal or abnormal. On the other hand, when the abnormal detection result output by the anomaly detection model indicates that the battery capacity is normal, the feature vectors of the manufacturing data representing the time series category of the battery and the feature vectors of the manufacturing data representing the discrete category of the battery are predicted and processed based on the regression model to predict the battery capacity. Thus, a data-driven battery capacity classification method is provided, which can predict the battery capacity of the battery with high precision by analyzing the manufacturing data of the lithium battery and combining the anomaly detection model and the regression model, thereby reducing the power consumption cost and time consumption, simplifying the entire manufacturing process of the battery, and improving the production efficiency and production safety of the battery.

[0023] Figure 1 This is an exemplary application scenario diagram provided by the embodiment of this application. In actual applications, the entire manufacturing process of lithium batteries usually includes: ingredient preparation process, coating process, baking process, lamination process, assembly process, liquid injection process, formation process and capacity separation process. Figure 1 In the application scenario shown, first, manufacturing data of one or more processes in the lithium battery manufacturing process is collected. Manufacturing data is relevant data for manufacturing lithium batteries, and can also be considered as production data of lithium batteries. Manufacturing data includes, but is not limited to: process data and manufacturing process data.

[0024] Process data can describe the characteristics of a process, including, but not limited to, the process name, equipment information required for the process, and process parameters. Process parameters refer to a series of basic parameter data required to complete a process, describing the task at hand. For example, process parameters for a batching process include raw material data. Process parameters for a liquid injection process include the electrolyte injection volume.

[0025] Among them, manufacturing process data refers to data generated during the manufacturing process of lithium batteries. Manufacturing process data may, for example, include process data generated by the lithium battery itself, such as the current, voltage, and shell temperature of the lithium battery at various moments during the formation process; manufacturing process data may, for example, include production process data of the process, such as the amount of raw materials used in the batching process and the amount of electrolyte consumed in the formation process.

[0026] In this embodiment, the collected manufacturing data is divided into time series manufacturing data and discrete manufacturing data. Time series manufacturing data is a type of time series data recorded in chronological order. For example, time series manufacturing data may include, but is not limited to: the current, voltage, and shell temperature of the lithium battery at various moments during the formation process. Discrete manufacturing data is a type of discrete data. For example, discrete manufacturing data may include, but is not limited to: raw material data of the batching process, the electrolyte injection volume of the injection process, and the electrolyte consumption of the formation process.

[0027] It is worth noting that in actual applications, the manufacturing data involved in battery capacity prediction can be flexibly selected as needed without any restrictions.

[0028] Secondly, data cleaning is performed on the manufacturing data of the time series category and the discrete category of lithium batteries, including but not limited to: data alignment, missing value processing, outlier processing, duplicate value removal and data conversion, etc.

[0029] Next, feature engineering is performed on the cleaned lithium battery manufacturing data of the time series category and the discrete category, obtaining feature vectors representing the manufacturing data of the time series category and the discrete category. For ease of understanding and distinction, the feature vectors representing the manufacturing data of the time series category are referred to as time series feature vectors, and the feature vectors representing the manufacturing data of the discrete category are referred to as discrete feature vectors. Feature engineering is the process of converting raw data into feature vectors that better express the essence of the problem. The purpose is to extract feature vectors from the raw data to the maximum extent possible for use in algorithms and models. During feature engineering, feature construction can be based on expert experience or on data statistics or machine learning models, but is not limited to these. For example, feature vectors constructed based on expert experience include, but are not limited to: the voltage of the lithium battery at the end of the formation process, the integral of the voltage of the lithium battery during the formation process over time, the integral of the current of the lithium battery during the formation process over time, the integral of the shell temperature of the lithium battery during the formation process over time, the amount of electrolyte injected during the injection process, the electrolyte consumption during the formation process, and so on. Feature vectors constructed based on statistical or machine learning models include, but are not limited to, the average voltage and current values, the gradient along the rising and falling edges of the voltage during the lithium battery formation process, or high-level features extracted by machine learning models from time-series or discrete manufacturing data. In practical applications, feature engineering can be flexibly performed as needed. For more information on feature engineering, please refer to the related technologies.

[0030] Finally, the AI ​​(Artificial Intelligence) model deployed on the server performs anomaly detection and high-precision prediction of the battery capacity of the lithium battery based on the time series feature vector and discrete feature vector of the lithium battery. Specifically, the AI ​​model includes an anomaly detection model and a regression model. First, the time series feature vector is input into the anomaly detection model to detect anomalies in the battery capacity of the lithium battery. If the battery capacity of the lithium battery is abnormal, the battery capacity of the lithium battery is no longer predicted. If the battery capacity of the lithium battery is normal, the time series feature vector and discrete feature vector are input into the regression model to predict the battery capacity of the lithium battery.

[0031] Compared with the existing method of lithium battery capacity classification based on the charging and discharging process, the data-driven lithium battery capacity classification method analyzes the manufacturing data of lithium batteries and combines anomaly detection models and regression models to predict the battery capacity of lithium batteries with high precision, reducing power consumption costs and time consumption, simplifying the entire lithium battery manufacturing process, and improving the production efficiency and production safety of lithium batteries.

[0032] It is worth noting that Figure 1 The application scenario shown is only an example, and the embodiments of the present application do not limit the application scenario.

[0033] The following describes in detail the technical solutions provided by various embodiments of the present application in conjunction with the accompanying drawings.

[0034] Figure 2 This is a flow chart of a battery capacity prediction method provided in an embodiment of the present application. The method can be executed by a battery capacity prediction device, which can be composed of hardware and / or software and can usually be integrated into an electronic device. Figure 2 , the method may include the following steps:

[0035] 201. Obtain manufacturing data of a target battery in a time series category and manufacturing data of a discrete category.

[0036] 202. Perform feature engineering processing on the manufacturing data of the time series category and the manufacturing data of the discrete category respectively to extract a first feature vector corresponding to the manufacturing data of the time series category and a second feature vector corresponding to the manufacturing data of the discrete category.

[0037] 203. Input the first eigenvector into an anomaly detection model to perform anomaly detection on the battery capacity of the target battery through the anomaly detection model.

[0038] 204. If the abnormality detection result indicates that the battery capacity of the target battery is normal, input the first eigenvector and the second eigenvector into a regression model to predict the battery capacity of the target battery through the regression model.

[0039] In this embodiment, the target battery refers to any battery that requires capacity sizing. Examples include, but are not limited to, lithium batteries, lead-acid batteries, and graphene batteries. Different battery types have different manufacturing processes, and the manufacturing processes of the same type of battery may vary between different manufacturers. In practical applications, the time-series manufacturing data and the discrete manufacturing data used for battery capacity prediction can be flexibly selected as needed, without limitation.

[0040] In practical applications, feature engineering can be directly performed on the manufacturing data of the time series category and the manufacturing data of the discrete category of the target battery. Further optionally, data cleaning is performed on the manufacturing data of the time series category and the manufacturing data of the discrete category of the target battery respectively, and then feature engineering is performed on the manufacturing data of the time series category and the manufacturing data of the discrete category of the target battery respectively. It can be understood that data cleaning can remove some dirty data, delete duplicate information, correct existing errors and ensure data consistency. Performing data cleaning before feature engineering can more accurately obtain the feature vectors representing the manufacturing data of the time series category and the feature vectors representing the manufacturing data of the discrete category. In order to facilitate understanding and distinction, the feature vector representing the manufacturing data of the time series category is called the first feature vector, and the feature vector representing the manufacturing data of the discrete category is called the second feature vector.

[0041] In this embodiment, an anomaly detection model for detecting anomalies in battery capacity is pre-trained. Anomaly detection models include, but are not limited to, convolutional neural networks (CNN), recurrent neural networks (RNN), and long short-term memory networks (LSTM). Further, optionally, in order to improve the recognition accuracy of the anomaly detection model, the anomaly detection model may be, for example, a generative adversarial network (GAN). The training process of the anomaly detection model will be described later.

[0042] In this embodiment, the first eigenvector of the target battery is input into the anomaly detection model to perform anomaly detection on the battery capacity of the target battery. If the anomaly detection result indicates that the battery capacity of the target battery is abnormal, the target battery is determined to be a battery that does not meet the battery capacity standard, and subsequent battery capacity prediction of the target battery is not required. If the anomaly detection result indicates that the battery capacity of the target battery is normal, the target battery is determined to be a battery that meets the battery capacity standard, and subsequent battery capacity prediction of the target battery can be performed.

[0043] In this embodiment, there is no restriction on the model structure of the anomaly detection model. Further, in order to improve the recognition accuracy of the anomaly detection model, see Figure 3 ,The anomaly detection model can include a generator for reconstruction ,processing and a discriminator for distinguishing whether the battery capacity ,is normal, which are connected in sequence. It is worth noting that the reconstruction ,processing can reconstruct a new feature vector reflecting the target ,battery.

[0044] Based on the above, a possible implementation method of inputting the first eigenvector into the anomaly detection model to perform anomaly detection on the battery capacity of the target battery through the anomaly detection model includes: using a generator to reconstruct the first eigenvector to obtain a reconstructed eigenvector corresponding to the first eigenvector; inputting the first eigenvector and the reconstructed eigenvector into the discriminator to perform anomaly detection on the battery capacity of the target battery through the discriminator.

[0045] In this embodiment, there is no restriction on the model structure of the generator. For example, the generator can be any autoencoder (AutoEncoder), which includes an encoder and a decoder. When the generator performs reconstruction processing, it first uses the encoder to perform encoding processing, and then uses the decoder to perform decoding processing. The decoding result is the reconstruction result.

[0046] Further optionally, in order to highlight the reconstruction error corresponding to the abnormal capacitance and reduce the reconstruction error corresponding to the normal capacitance, a memory module (Memory Module) for data enhancement processing can be added to the generator. The memory module is used to store a number of enhancement parameters, and the memory module uses these enhancement parameters to perform data enhancement processing on the encoding result output by the encoder. For example, the memory module stores a weight matrix, which includes multiple weights, which are enhancement parameters. The memory module uses each weight in the weight matrix to perform weighted summation on the encoding result output by the encoder.

[0047] Based on the above, as an example, see Figure 3 The generator includes a first encoder with an encoding function, a memory module, and a first decoder with a decoding function, which are connected in sequence. Accordingly, the generator is used to reconstruct the first eigenvector to obtain a reconstructed eigenvector corresponding to the first eigenvector, including: encoding the first eigenvector with the first encoder to obtain a first encoding vector; performing a matrix transformation on the first encoding vector to obtain a transformed first encoding vector; performing data enhancement processing on the transformed first encoding vector with the memory module to obtain a data-enhanced first encoding vector; and decoding the data-enhanced first encoding vector with the decoder to obtain a reconstructed eigenvector corresponding to the first eigenvector.

[0048] In this embodiment, matrix transformation can change the dimensionality of the code vector. Matrix transformation aimed at dimensionality reduction can be understood as dimensionality reduction processing. Assume that the dimensionality of the first code vector is M×N, that is, the first code vector is a matrix with M rows and N columns, where M and N are positive integers; after the first code vector undergoes matrix transformation, the dimensionality of the transformed first code vector is (M×N)×1, that is, the transformed first code vector is a matrix with (M×N) rows and 1 column.

[0049] In this embodiment, there is no restriction on the model structure of the discriminator. As an example, the discriminator includes a second encoder with an encoding function and a pooling layer with a pooling function connected in sequence; accordingly, the first feature vector and the reconstructed feature vector are input into the discriminator, and an optional implementation method for performing abnormality detection on the battery capacity of the target battery through the discriminator is: using the second encoder to encode the first feature vector and the reconstructed feature vector to obtain a second encoding vector; using the pooling layer to pool the second encoding vector to obtain a pooling result; using the activation function to activate the pooling result to obtain an abnormality detection result for the battery capacity of the target battery.

[0050] In this embodiment, there is no restriction on the activation function, and the activation function includes, but is not limited to, a Sigmoid activation function, a hyperbolic tangent activation function, and a ReLU activation function.

[0051] In this embodiment, when the abnormality detection result output by the abnormality detection model indicates that the battery capacity of the target battery is normal, the battery capacity of the target battery is further predicted using the regression model. Regression models include, but are not limited to, various classification algorithm models such as logistic regression model (LogisticRegression), K-nearest neighbor classification (KNN), support vector machine (SVM), and decision tree, and are not limited to these.

[0052] In practical applications, the first eigenvector and the second eigenvector can be directly input into the regression model to predict the battery capacity of the target battery through the regression model. Further, in order to improve the prediction accuracy of the regression model, see Figure 4, a third encoder with encoding functionality can be used to encode the first eigenvector to obtain a third coded vector; the third coded vector and the second eigenvector are concatenated to obtain a concatenated vector; the concatenated vector is input into a regression model to predict the target battery capacity using the regression model. In practical applications, if the third coded vector and the second eigenvector have different dimensional information, the third coded vector can be first subjected to a matrix transformation to equalize the dimensional information of the transformed third coded vector and the second eigenvector, and then the concatenated vector is concatenated.

[0053] In some optional embodiments, to improve the prediction accuracy of the regression model, the residual between the feature vector representing the manufacturing data of the time series category and its corresponding reconstructed feature vector may also be considered. As an example, feature concatenating the third encoded vector and the second feature vector to obtain a concatenated vector includes: encoding the reconstructed feature vector of the first feature vector using a third encoder to obtain a fourth encoded vector; determining a target residual between the third encoded vector and the fourth encoded vector; and feature concatenating the third encoded vector, the target residual, and the second feature vector to obtain a concatenated vector.

[0054] For further options, see Figure 4 To improve the accuracy of the residual, the third and fourth code vectors can be processed separately using fully connected layers (FCs). The target residual between the third and fourth code vectors processed by the FCs is then determined. In practical applications, if the fourth eigenvector and the second eigenvector have different dimensional information, the fourth code vector can be transformed by a matrix to equalize the dimensional information of the transformed fourth and second eigenvectors. The transformed fourth code vector can then be input into the FCs for processing.

[0055] The technical solution provided by the embodiments of the present application, on the one hand, processes the feature vectors of the manufacturing data representing the time series categories of the battery based on an anomaly detection model to detect whether the battery capacity is normal or abnormal. On the other hand, when the anomaly detection result output by the anomaly detection model indicates that the battery capacity is normal, the feature vectors of the manufacturing data representing the time series categories of the battery and the feature vectors of the manufacturing data representing the discrete categories of the battery are predicted and processed based on a regression model to predict the battery capacity. Thus, a data-driven battery capacity classification method is provided, which can accurately predict the battery capacity of the battery, reduce power consumption costs and time consumption, simplify the entire battery manufacturing process, and improve battery production efficiency and production safety.

[0056] The following introduces the training process of the anomaly detection model and the regression model respectively.

[0057] The present invention provides a model training method, which may include the following steps:

[0058] 11. Obtain sample time series feature vectors of multiple sample batteries.

[0059] 12. Perform model training on the initial anomaly detection model according to the sample feature vectors of multiple sample batteries to obtain an anomaly detection model.

[0060] In this embodiment, manufacturing data of the time series category is first collected for multiple sample batteries. These sample batteries are batteries with normal battery capacity. It is worth noting that if the battery capacity falls within the set capacity range, it can be considered normal; if the battery capacity does not fall within the set capacity range, it can be considered abnormal.

[0061] Next, feature engineering is performed on the manufacturing data of the sample battery's time series category to obtain a feature vector reflecting the manufacturing data of the sample battery's time series category. For ease of understanding and distinction, the feature vector of the manufacturing data of the sample battery reflecting the time series category is referred to as the sample time series feature vector. Optionally, data cleaning can be performed before feature engineering.

[0062] Finally, after obtaining the sample time series feature vectors of multiple sample batteries, the initial anomaly detection model is trained according to the sample feature vectors of the multiple sample batteries to obtain an anomaly detection model. In practical applications, the sample time series feature vectors of multiple sample batteries can be used for multiple model trainings. After each model training, the anomaly detection model after the model parameters are adjusted is used as the next anomaly detection model to be trained, and the next model training is performed using the new sample time series feature vectors of multiple sample batteries until the end training condition is met, and the anomaly detection model obtained from the last training is used as the final anomaly detection model. Among them, the end training condition can be that the number of model training times reaches a specified number, or that the anomaly detection model converges, and there is no specific restriction here.

[0063] The present embodiment does not limit the training method of the anomaly detection model. The following describes several optional training methods:

[0064] Method 1: For each sample battery, the sample time series feature vector of the sample battery is input into the generator of the initial anomaly detection model to obtain the sample reconstruction feature vector output by the generator; the sample time series feature vector and the sample reconstruction feature vector are input into the discriminator of the anomaly detection model to obtain the anomaly detection result output by the discriminator, and the anomaly detection result reflects the probability that the battery capacity of the sample battery is abnormal; the reconstruction loss of the sample battery is determined based on the sample time series feature vector and the sample reconstruction feature vector; the model parameters of the generator are adjusted according to the reconstruction loss of each sample battery, and the model parameters of the discriminator are adjusted according to the anomaly detection results of each sample battery to obtain an anomaly detection model.

[0065] Method 2: For each sample battery, the sample time series feature vector of the sample battery is input into the generator of the initial anomaly detection model to obtain the sample reconstruction feature vector output by the generator; the sample time series feature vector and the sample reconstruction feature vector are input into the discriminator of the anomaly detection model to obtain the anomaly detection result output by the discriminator, and the anomaly detection result reflects the probability that the battery capacity of the sample battery is abnormal; the encoded sample time series feature vector output by the first encoder in the generator is obtained; the encoded sample time series feature vector output by the second encoder in the discriminator is obtained; the reconstruction loss of the sample battery is determined based on the sample time series feature vector and the sample reconstruction feature vector; the adversarial loss of the sample battery is determined based on the encoded sample time series feature vectors output by the first encoder and the second encoder respectively; the model parameters of the generator are adjusted according to the reconstruction loss and adversarial loss of each sample battery, and the model parameters of the discriminator are adjusted according to the anomaly detection results of each sample battery to obtain an anomaly detection model.

[0066] It is worth noting that in the model training process, method 2 also considers adversarial loss. The recognition accuracy of the anomaly detection model trained by method 2 is better than that of the anomaly detection model trained by method 1.

[0067] In this embodiment, the reconstruction loss or adversarial loss includes, but is not limited to: a logarithmic loss function, an L1 distance loss (L1 Loss) loss function, a cross-entropy loss function, and a Focal loss loss function for solving data imbalance problems.

[0068] It is worth noting that when the reconstruction loss and / or adversarial loss of each sample battery adjusts the model parameters of the generator, the reconstruction loss and / or adversarial loss of multiple sample batteries can be subjected to various numerical calculations such as summation, average or accumulation to obtain the target loss function, and the target loss function can be used to adjust the model parameters of the anomaly detection model.

[0069] It's worth noting that the model parameters of the generator and discriminator can be adjusted synchronously or asynchronously, with no restrictions. Synchronous adjustment can be understood as adjusting the model parameters of both the generator and discriminator simultaneously during each round of model training. Asynchronous adjustment can be understood as adjusting only the model parameters of the generator or discriminator during a round of model training, with the model parameters of the generator and discriminator adjusted separately in different rounds of model training.

[0070] It is worth noting that the anomaly detection model provided in this embodiment can accurately detect anomalies in the battery capacity of a lithium battery. The anomaly detection model, combined with the regression model, can predict the battery capacity of a lithium battery with high precision, thereby reducing power consumption costs and time consumption, simplifying the entire manufacturing process of the lithium battery, and improving the production efficiency and safety of the lithium battery.

[0071] The present invention also provides another model training method, which may include the following steps:

[0072] 21. Obtain sample time series feature vectors, sample discrete feature vectors and expected battery capacities of multiple sample batteries. The battery capacities of the sample batteries are normal. The sample time series feature vectors are obtained by performing feature engineering processing on the manufacturing data of the time series category of the sample batteries. The sample discrete feature vectors are obtained by performing feature engineering processing on the manufacturing data of the discrete category of the sample batteries.

[0073] 22. Using the sample time series feature vectors, sample discrete feature vectors and expected battery capacity of multiple sample batteries, the initial regression model is trained to obtain a regression model.

[0074] In this embodiment, when training the regression model, in addition to collecting the manufacturing data of multiple sample batteries in time series categories, it is also necessary to collect the manufacturing data of multiple sample batteries in discrete categories, and label the expected battery capacity of each of the multiple sample batteries.

[0075] First, feature engineering is performed on the manufacturing data of the sample batteries in the time series category and the discrete category, respectively, to obtain sample time series feature vectors reflecting the manufacturing data in the time series category and sample discrete feature vectors reflecting the manufacturing data in the discrete category. Next, the initial regression model is trained using the sample time series feature vectors, sample discrete feature vectors, and expected battery capacity of multiple sample batteries to obtain a regression model.

[0076] The present embodiment does not limit the training method of the regression model. The following are several optional training methods:

[0077] Method 1: For each sample battery, the sample discrete feature vector and the sample time series feature vector are concatenated to obtain a sample concatenated vector; the sample concatenated vector is input into the initial regression model to obtain the predicted battery capacity of the sample battery; according to the predicted battery capacity and expected battery capacity of each sample battery, the model parameters of the initial regression model are adjusted to obtain a regression model.

[0078] Method 2: For each sample battery, use a third encoder to encode the sample time series feature vector to obtain an encoded sample time series feature vector; perform feature splicing on the sample discrete feature vector and the encoded sample time series feature vector to obtain a sample splicing vector; input the sample splicing vector into the initial regression model to obtain the predicted battery capacity of the sample battery; adjust the model parameters of the initial regression model according to the predicted battery capacity and expected battery capacity of each sample battery to obtain a regression model.

[0079] Method 3: For each sample battery, use the third encoder to encode the sample time series feature vector to obtain the encoded sample time series feature vector; use the third encoder to encode the sample reconstruction feature vector corresponding to the sample time series feature vector to obtain the encoded sample reconstruction feature vector; determine the sample residual between the sample time series feature vector and the sample reconstruction feature vector; perform feature splicing on the sample discrete feature vector, the sample residual and the encoded sample time series feature vector to obtain a sample splicing vector; input the sample splicing vector into the initial regression model to obtain the predicted battery capacity of the sample battery; adjust the model parameters of the initial regression model according to the predicted battery capacity and the expected battery capacity of each sample battery to obtain a regression model.

[0080] It is worth noting that the sample reconstruction feature vector corresponding to the sample time series feature vector can be obtained using the anomaly detection model. The sample residual refers to the residual between the sample time series feature vector and the sample reconstruction feature vector. The method for determining the residual can refer to the introduction to the residual in the above content.

[0081] It is worth noting that the order of prediction accuracy from high to low is: regression model trained by method 3, regression model trained by method 2, and regression model trained by method 1.

[0082] In this embodiment, when adjusting the model parameters of the initial regression model based on the predicted battery capacity and expected battery capacity of each sample battery, the loss function corresponding to each sample battery is first calculated based on the predicted battery capacity and expected battery capacity of each sample battery. The loss function includes, but is not limited to, a logarithmic loss function, an L1 distance loss (L1Loss) loss function, a cross-entropy loss function, and a focal loss function for solving data imbalance problems. Subsequently, various numerical calculations such as summation, averaging, or accumulation can be performed on the loss functions of the multiple sample batteries to obtain a target loss function, which is then used to adjust the model parameters of the regression model.

[0083] In practical applications, regression models can be trained multiple times. After each training session, the regression model with adjusted model parameters is used as the next regression model to be trained. The next training session is performed using multiple new sample data until the training termination criteria are met. The regression model obtained from the last training session is then used as the final regression model. The termination criteria can be reaching a specified number of training cycles or convergence of the regression model, which are not specifically limited here.

[0084] It is worth noting that the regression model provided in this embodiment is combined with the anomaly detection model to predict the battery capacity of the lithium battery with high precision, reducing the power consumption cost and time consumption, simplifying the entire manufacturing process of the lithium battery, and improving the production efficiency and production safety of the lithium battery.

[0085] Figure 6 This is a schematic diagram of the structure of a battery capacity prediction device provided in an embodiment of the present application. Figure 6 , the apparatus may include:

[0086] An acquisition module 61 is configured to acquire time-series manufacturing data and discrete manufacturing data of a target battery, wherein the time-series manufacturing data represents data generated during the manufacturing process of the target battery, and the discrete manufacturing data represents data of the manufacturing process of the target battery;

[0087] A feature engineering module 62 is configured to perform feature engineering processing on the time series manufacturing data and the discrete manufacturing data, respectively, to extract a first feature vector corresponding to the time series manufacturing data and a second feature vector corresponding to the discrete manufacturing data;

[0088] an anomaly detection module 63, configured to input the first feature vector into an anomaly detection model to perform an anomaly detection on the battery capacity of the target battery through the anomaly detection model;

[0089] The battery capacity prediction module 64 is configured to input the first eigenvector and the second eigenvector into a regression model if the abnormality detection result indicates that the battery capacity of the target battery is normal, so as to predict the battery capacity of the target battery through the regression model.

[0090] Further optionally, the anomaly detection model includes a generator and a discriminator connected in sequence; accordingly, when the anomaly detection module 63 inputs the first feature vector into the anomaly detection model to perform anomaly detection on the battery capacity of the target battery through the anomaly detection model, it is specifically used to: use the generator to reconstruct the first feature vector to obtain a reconstructed feature vector corresponding to the first feature vector; input the first feature vector and the reconstructed feature vector into the discriminator to perform anomaly detection on the battery capacity of the target battery through the discriminator.

[0091] Further optionally, the generator includes a first encoder, a memory module and a first decoder connected in sequence. Accordingly, when the anomaly detection module 63 uses the generator to reconstruct the first eigenvector to obtain a reconstructed eigenvector corresponding to the first eigenvector, it is specifically used to: use the first encoder to encode the first eigenvector to obtain a first encoding vector; perform matrix transformation on the first encoding vector to obtain a transformed first encoding vector; use the memory module to perform data enhancement processing on the transformed first encoding vector to obtain a data-enhanced first encoding vector; use the decoder to decode the data-enhanced first encoding vector to obtain a reconstructed eigenvector corresponding to the first eigenvector.

[0092] Further optionally, the discriminator includes a second encoder and a pooling layer connected in sequence; accordingly, when the anomaly detection module 63 inputs the first feature vector and the reconstructed feature vector into the discriminator to perform anomaly detection on the battery capacity of the target battery through the discriminator, it is specifically used to: use the second encoder to encode the first feature vector and the reconstructed feature vector to obtain a second encoding vector; use the pooling layer to pool the second encoding vector to obtain a pooling result; use the activation function to activate the pooling result to obtain an anomaly detection result for the battery capacity of the target battery.

[0093] Further optionally, when the battery capacity prediction module 64 inputs the first eigenvector and the second eigenvector into the regression model to predict the battery capacity of the target battery through the regression model, it is specifically used to: encode the first eigenvector using a third encoder to obtain a third encoded vector; perform feature splicing on the third encoded vector and the second eigenvector to obtain a spliced ​​vector; and input the spliced ​​vector into the regression model to predict the battery capacity of the target battery through the regression model.

[0094] Further optionally, the battery capacity prediction module 64 performs feature splicing on the third coding vector and the second eigenvector to obtain a spliced ​​vector, which is specifically used to: use the third encoder to encode the reconstructed feature vector of the first eigenvector to obtain a fourth coding vector; determine the target residual between the third coding vector and the fourth coding vector; and perform feature splicing on the third coding vector, the target residual, and the second eigenvector to obtain a spliced ​​vector.

[0095] Further optionally, when the battery capacity prediction module 64 determines the target residual between the third coding vector and the fourth coding vector, it is specifically used to: use the fully connected layer to process the third coding vector and the fourth coding vector respectively; and determine the target residual between the third coding vector and the fourth coding vector after processing by the fully connected layer.

[0096] Further optionally, the above-mentioned device also includes a training module for obtaining sample time series feature vectors, sample discrete feature vectors and expected battery capacity of multiple sample batteries, the battery capacity of the sample batteries is normal, the sample time series feature vectors are obtained by feature engineering processing of the manufacturing data of the time series category of the sample batteries, and the sample discrete feature vectors are obtained by feature engineering processing of the manufacturing data of the discrete category of the sample batteries; the sample time series feature vectors, sample discrete feature vectors and expected battery capacity of multiple sample batteries are used to train the initial regression model to obtain a regression model.

[0097] Further optionally, the training module uses the sample time series feature vectors, sample discrete feature vectors and expected battery capacities of multiple sample batteries to train the initial regression model. When the regression model is obtained, it is specifically used to: for each sample battery, use a third encoder to encode the sample time series feature vector to obtain the encoded sample time series feature vector; feature splicing the sample discrete feature vector and the encoded sample time series feature vector to obtain a sample splicing vector; input the sample splicing vector into the initial regression model to obtain the predicted battery capacity of the sample battery; adjust the model parameters of the initial regression model according to the predicted battery capacity and the expected battery capacity of each sample battery to obtain a regression model.

[0098] Further optionally, the training module is further used to: obtain sample time series feature vectors of multiple sample batteries; and perform model training on the initial anomaly detection model according to the sample feature vectors of the multiple sample batteries to obtain the anomaly detection model.

[0099] Further optionally, the training module performs model training on the initial anomaly detection model based on the sample feature vectors of multiple sample batteries. When the anomaly detection model is obtained, it is specifically used to: for each sample battery, input the sample time series feature vector of the sample battery into the generator of the initial anomaly detection model to obtain the sample reconstruction feature vector output by the generator; input the sample time series feature vector and the sample reconstruction feature vector into the discriminator of the anomaly detection model to obtain the anomaly detection result output by the discriminator, and the anomaly detection result reflects the probability that the battery capacity of the sample battery is abnormal; determine the reconstruction loss of the sample battery based on the sample time series feature vector and the sample reconstruction feature vector; adjust the model parameters of the generator according to the reconstruction loss of each sample battery, and adjust the model parameters of the discriminator according to the anomaly detection result of each sample battery to obtain the anomaly detection model.

[0100] Further optionally, the training module is also used to: obtain the encoded sample time series feature vector output by the first encoder in the generator; obtain the encoded sample time series feature vector output by the second encoder in the discriminator; determine the adversarial loss of the sample battery based on the encoded sample time series feature vectors output by the first encoder and the second encoder respectively; and adjust the model parameters of the generator according to the adversarial loss of each sample battery.

[0101] Figure 6 The device shown can perform Figure 2 The method shown in FIG. 1 and its implementation principle and technical effects are not described in detail. Figure 6 The specific manner in which each module and unit performs operations in the device shown has been described in detail in the embodiment of the method and will not be elaborated here.

[0102] It should be noted that the execution entity of each step of the method provided in the above embodiment can be the same device, or the method can be executed by different devices. For example, the execution entity of steps 201 to 204 can be device A; for another example, the execution entity of steps 201 and 202 can be device A, and the execution entity of steps 203 and 204 can be device B; and so on.

[0103] In addition, in some of the processes described in the above embodiments and the accompanying drawings, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations, such as 201, 202, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0104] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 7 As shown, the electronic device includes: a memory 71 and a processor 72;

[0105] The memory 71 is used to store computer programs and can be configured to store various other data to support operations on the computing platform. Examples of such data include instructions for any application or method operating on the computing platform, contact data, phone book data, messages, pictures, videos, etc.

[0106] The memory 71 can be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read only memory (EEPROM), erasable programmable read only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0107] The processor 72 is coupled to the memory 71 and is configured to execute a computer program in the memory 71 to: obtain manufacturing data of a time series category and manufacturing data of a discrete category of a target battery, wherein the manufacturing data of the time series category represents data generated during the manufacturing process of the target battery, and the manufacturing data of the discrete category represents data of the manufacturing process of the target battery; perform feature engineering processing on the manufacturing data of the time series category and the manufacturing data of the discrete category, respectively, to extract a first feature vector corresponding to the manufacturing data of the time series category and a second feature vector corresponding to the manufacturing data of the discrete category; input the first feature vector into an anomaly detection model to perform anomaly detection on the battery capacity of the target battery through the anomaly detection model; if the anomaly detection result indicates that the battery capacity of the target battery is normal, input the first feature vector and the second feature vector into a regression model to predict the battery capacity of the target battery through the regression model;

[0108] Alternatively, the processor 72 is coupled to the memory 71 and is configured to execute a computer program in the memory 71 to: obtain sample time series feature vectors of a plurality of sample batteries; perform model training on an initial anomaly detection model based on the sample feature vectors of the plurality of sample batteries to obtain an anomaly detection model;

[0109] Alternatively, the processor 72 is coupled to the memory 71 and is configured to execute a computer program in the memory 71 to: obtain sample time series feature vectors, sample discrete feature vectors, and expected battery capacities of multiple sample batteries, wherein the battery capacities of the sample batteries are normal, the sample time series feature vectors are obtained by subjecting the manufacturing data of the time series categories of the sample batteries to feature engineering, and the sample discrete feature vectors are obtained by subjecting the manufacturing data of the discrete categories of the sample batteries to feature engineering; and train an initial regression model using the sample time series feature vectors, sample discrete feature vectors, and expected battery capacities of the multiple sample batteries to obtain a regression model. Further, as Figure 7 As shown, the electronic device also includes: a communication component 73, a display 74, a power component 75, an audio component 76 and other components. Figure 7 Only some components are shown schematically, which does not mean that the electronic device only includes Figure 7 In addition, Figure 7 The components in the dotted box are optional components, not mandatory components, and the specific components may depend on the product form of the electronic device. The electronic device of this embodiment can be implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone or an IOT (Internet of Things) device, or a server device such as a conventional server, a cloud server or a server array. If the electronic device of this embodiment is implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone, etc., it can include Figure 7If the electronic device of this embodiment is implemented as a conventional server, cloud server or server array and other server-side devices, it may not include Figure 7 Components within the dotted box.

[0110] The detailed implementation process of the processor executing each action can be found in the relevant description in the aforementioned method embodiment or device embodiment, and will not be repeated here.

[0111] Accordingly, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed, can implement the steps that can be performed by the electronic device in the above method embodiment.

[0112] Accordingly, an embodiment of the present application also provides a computer program product, including a computer program / instruction. When the computer program / instruction is executed by a processor, the processor is enabled to implement the steps in the above method embodiment that can be performed by an electronic device.

[0113] The above-mentioned communication component is configured to facilitate wired or wireless communication between the device where the communication component is located and other devices. The device where the communication component is located can access a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G / LTE, 5G and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra wide band (UWB) technology, Bluetooth (BT) technology and other technologies.

[0114] The above-mentioned display includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor can not only sense the boundary of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation.

[0115] The power supply assembly provides power to various components of the device in which the power supply assembly is located. The power supply assembly may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which the power supply assembly is located.

[0116] The above-mentioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC), and when the device where the audio component is located is in an operating mode, such as call mode, recording mode, and voice recognition mode, the microphone is configured to receive external audio signals. The received audio signal can be further stored in a memory or sent via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0117] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can 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 can adopt the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0118] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes 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 produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0119] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work 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 The function specified in one or more boxes.

[0120] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device 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.

[0121] In a typical configuration, a computing device includes one or more processors (central processing units, CPUs), input / output interfaces, network interfaces, and memory.

[0122] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0123] Computer-readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, Phase Change RAM (PRAM), Static Random-Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically-Erasable Programmable Read-Only Memory (EEPROM), flash memory or other memory technology, CD-ROM, Digital Versatile Disc (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device. According to the definition in this article, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0124] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0125] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A battery capacity prediction method, characterized in that: include: Acquire manufacturing data of a time series category and manufacturing data of a discrete category of a target battery; Performing feature engineering processing on the manufacturing data of the time series category and the manufacturing data of the discrete category respectively to extract a first feature vector corresponding to the manufacturing data of the time series category and a second feature vector corresponding to the manufacturing data of the discrete category; Inputting the first feature vector into an anomaly detection model, the anomaly detection model comprising a generator and a discriminator connected in sequence, reconstructing the first feature vector using the generator to obtain a reconstructed feature vector corresponding to the first feature vector; inputting the first feature vector and the reconstructed feature vector into the discriminator, so as to perform an anomaly detection on the battery capacity of the target battery through the discriminator; If the abnormality detection result indicates that the battery capacity of the target battery is normal, the first feature vector and the second feature vector are input into a regression model to predict the battery capacity of the target battery through the regression model.

2. The method according to claim 1, characterized in that The generator includes a first encoder, a memory module, and a first decoder connected in sequence. Accordingly, the generator is used to reconstruct the first feature vector to obtain a reconstructed feature vector corresponding to the first feature vector, including: Encoding the first feature vector using the first encoder to obtain a first encoding vector; Performing a matrix transformation on the first code vector with the goal of reducing dimensionality to obtain a transformed first code vector; Performing data enhancement processing on the transformed first coding vector using the memory module to obtain a data-enhanced first coding vector; The decoder is used to decode the first encoded vector after data enhancement to obtain a reconstructed feature vector corresponding to the first feature vector.

3. The method according to claim 1, characterized in that The discriminator includes a second encoder and a pooling layer connected in sequence; accordingly, the first feature vector and the reconstructed feature vector are input into the discriminator to perform abnormality detection on the battery capacity of the target battery through the discriminator, including: Using the second encoder to encode the first feature vector and the reconstructed feature vector to obtain a second encoded vector; Performing pooling processing on the second encoding vector using the pooling layer to obtain a pooling result; The pooling result is activated by using an activation function to obtain an abnormality detection result of the battery capacity of the target battery.

4. The method according to claim 1, wherein Inputting the first eigenvector and the second eigenvector into a regression model to predict the battery capacity of the target battery through the regression model includes: Encoding the first feature vector using a third encoder to obtain a third encoded vector; Perform feature splicing on the third encoding vector and the second feature vector to obtain a spliced ​​vector; The concatenated vector is input into the regression model to predict the battery capacity of the target battery through the regression model.

5. The method according to claim 4, characterized in that Performing feature splicing on the third encoding vector and the second feature vector to obtain a spliced ​​vector includes: encoding the reconstructed feature vector of the first feature vector using a third encoder to obtain a fourth encoded vector; determining a target residual between the third code vector and the fourth code vector; The third encoding vector, the target residual, and the second feature vector are feature-concatenated to obtain the concatenated vector.

6. The method according to claim 5, characterized in that Determining a target residual between the third code vector and the fourth code vector includes: Processing the third encoding vector and the fourth encoding vector respectively using a fully connected layer; Determine a target residual between the third encoding vector and the fourth encoding vector after being processed by the fully connected layer.

7. A model training method, characterized in that: include: Obtaining sample time series feature vectors, sample discrete feature vectors, and expected battery capacities of multiple sample batteries, where the battery capacities of the sample batteries are normal, the sample time series feature vectors being obtained by performing feature engineering processing on manufacturing data of a time series category of the sample batteries, and the sample discrete feature vectors being obtained by performing feature engineering processing on manufacturing data of a discrete category of the sample batteries; The initial regression model is trained using the sample time series feature vectors, sample discrete feature vectors and expected battery capacities of multiple sample batteries to obtain the regression model, wherein the regression model is used to predict the battery capacity of the target battery when executing the battery capacity prediction method according to any one of claims 1 to 6.

8. The method according to claim 7, characterized in that The initial regression model is trained using the sample time series feature vectors, sample discrete feature vectors, and expected battery capacities of a plurality of sample batteries to obtain the regression model, including: For each sample battery, use a third encoder to encode the sample time series feature vector to obtain an encoded sample time series feature vector; Perform feature splicing on the sample discrete feature vector and the encoded sample time series feature vector to obtain a sample splicing vector; Inputting the sample splicing vector into the initial regression model to obtain the predicted battery capacity of the sample battery; According to the predicted battery capacity and the expected battery capacity of each of the sample batteries, the model parameters of the initial regression model are adjusted to obtain the regression model.

9. A model training method, characterized in that: include: Obtaining sample time series feature vectors of multiple sample batteries; An initial anomaly detection model is trained based on sample feature vectors of multiple sample batteries to obtain an anomaly detection model; wherein the anomaly detection model is used to perform anomaly detection on the battery capacity of the target battery when executing the battery capacity prediction method according to any one of claims 1 to 6.

10. The method according to claim 9, characterized in that The initial anomaly detection model is trained based on the sample feature vectors of multiple sample batteries to obtain an anomaly detection model, including: For each sample battery, input the sample time series feature vector of the sample battery into the generator of the initial anomaly detection model to obtain the sample reconstruction feature vector output by the generator; Inputting the sample time series feature vector and the sample reconstruction feature vector into the discriminator of the anomaly detection model, obtaining an anomaly detection result output by the discriminator, wherein the anomaly detection result reflects the probability that the battery capacity of the sample battery is abnormal; Determining the reconstruction loss of the sample battery according to the sample time series feature vector and the sample reconstruction feature vector; The model parameters of the generator are adjusted according to the reconstruction loss of each sample battery, and the model parameters of the discriminator are adjusted according to the anomaly detection result of each sample battery to obtain the anomaly detection model.

11. The method according to claim 10, characterized in that Also includes: Obtaining an encoded sample time series feature vector output by the first encoder in the generator; Obtaining an encoded sample time series feature vector output by a second encoder in the discriminator; Determining the adversarial loss of the sample battery according to the encoded sample time series feature vectors output by the first encoder and the second encoder; The model parameters of the generator are adjusted according to the adversarial loss of each sample battery.

12. An electronic device, characterized in that: include: memory and processor; The memory is used to store computer programs; The processor is coupled to the memory and configured to execute the computer program to perform the steps of the method according to any one of claims 1 to 11.

13. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor is enabled to implement the steps of the method according to any one of claims 1 to 11.

Images