Lithium precipitation detection model training method based on semi-supervised deep clustering and lithium precipitation detection method

Through the semi-supervised deep clustering method, the lithium-ion detection model is trained, and the operating data and initial pseudo-label of the lithium battery are used to solve the problem of low diagnostic accuracy of lithium-ion faults in the existing technology, and the accuracy of identification of lithium-ion phenomena in lithium-ion batteries is improved. and model stability.

CN118940068AInactive Publication Date: 2024-11-12ELECTRIC POWER SCI & RES INST OF STATE GRID TIANJIN ELECTRIC POWER CO +2

Patent Information

Application Number
CN202411426942.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-14
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Among the existing lithium-ion battery fault diagnosis methods, the non-destructive detection method based on voltage data is limited, making it difficult to fully reflect the changes in the internal state of the battery. The unsupervised learning method lacks clear label information, resulting in low diagnostic accuracy and reliability. .

Method used

The lithium-excitation detection model training method based on semi-supervised deep clustering is adopted to obtain the operating data of multiple lithium batteries, including the first type of lithium-excitation type label and the second type of lithium-excitation type without labels, and the feature extraction is performed. and semi-supervised training to generate lithium-ion detection models, and use initial pseudo-labels and supervised data for model optimization.

Benefits of technology

Reliance on label data is reduced, the cost and difficulty of data annotation is reduced, the robustness and stability of lithium-ion detection model is improved, and the accuracy of identification of lithium-ion phenomena in lithium batteries is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940068A_ABST
    Figure CN118940068A_ABST
Patent Text Reader

Abstract

The invention provides a training method of a lithium precipitation detection model based on semi-supervised deep clustering and a lithium precipitation detection method, and mainly relates to the technical field of battery fault detection. The training method of the lithium precipitation detection model based on semi-supervised deep clustering comprises the steps that respective operation data of a plurality of lithium batteries are acquired, and the plurality of lithium batteries comprise a plurality of first-class lithium batteries with lithium precipitation class labels and a plurality of second-class lithium batteries without labels; performing feature extraction on the respective operation data of the plurality of lithium batteries to obtain respective first feature data of the plurality of lithium batteries; and performing semi-supervised training on the initial model based on the respective first feature data of the plurality of lithium batteries and the respective lithium precipitation category labels of the plurality of first-class lithium batteries to obtain a lithium precipitation detection model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of battery fault detection, and more specifically, to a training method for a lithium deposition detection model based on semi-supervised deep clustering and a lithium deposition detection method. Background Art

[0002] With the continuous advancement of technology and the increasing complexity of application scenarios, lithium-ion batteries are facing more stringent operating conditions. Lithium precipitation is one of the important causes of lithium-ion battery failure. It not only reduces the capacity of the battery, but the irreversible accumulation of lithium may also form lithium dendrites, pierce the diaphragm and cause internal short circuits in the battery, which in turn leads to serious safety problems such as thermal runaway.

[0003] At present, the diagnostic methods for lithium plating failures in lithium-ion batteries mainly include non-destructive testing methods based on voltage data and unsupervised learning methods based on multi-dimensional feature clustering. Among them, the non-destructive testing method is limited by the limited amount of information contained in the voltage data, and it is difficult to fully reflect the internal state changes of the battery, which easily affects the diagnostic accuracy and reliability. The unsupervised learning method captures subtle changes in the battery state through multi-dimensional features. However, due to the lack of clear label information in unsupervised learning, the training and evaluation of the model face great challenges, resulting in low accuracy of the diagnostic results. Summary of the invention

[0004] In view of this, the present invention provides a training method for a lithium deposition detection model based on semi-supervised deep clustering and a lithium deposition detection method.

[0005] An embodiment of the present invention provides a training method for a lithium plating detection model based on semi-supervised deep clustering, the method comprising: obtaining operating data of each of a plurality of lithium batteries, the plurality of lithium batteries comprising a plurality of first-category lithium batteries with lithium plating category labels and a plurality of second-category lithium batteries without labels; performing feature extraction on the operating data of each of the plurality of lithium batteries respectively to obtain first feature data of each of the plurality of lithium batteries; performing semi-supervised training on an initial model based on the first feature data of each of the plurality of lithium batteries and the lithium plating category labels of each of the plurality of first-category lithium batteries to obtain a lithium plating detection model.

[0006] According to an embodiment of the present invention, the above-mentioned semi-supervised training is performed on the initial model based on the first feature data of each of the above-mentioned multiple lithium batteries and the lithium plating category labels of each of the above-mentioned first-category lithium batteries to obtain a lithium plating detection model, including: clustering processing is performed based on the first feature data of each of the above-mentioned multiple second-category lithium batteries to determine the initial pseudo-labels of each of the above-mentioned multiple second-category lithium batteries; and the above-mentioned initial model is trained using the first feature data of each of the above-mentioned multiple first-category lithium batteries with lithium plating category labels and the first feature data of each of the multiple second-category lithium batteries with initial pseudo-labels to obtain the above-mentioned lithium plating detection model.

[0007] According to an embodiment of the present invention, the above-mentioned initial model is trained using the first feature data of each of the multiple first-type lithium batteries with lithium plating category labels and the first feature data of each of the multiple second-type lithium batteries with initial pseudo labels to obtain the above-mentioned lithium plating detection model, including: in the i-th training round, a plurality of training samples are determined from the first feature data of each of the multiple lithium batteries, the multiple training samples include at least one first training data and at least one second training data, the first training data is the first feature data of the first-type lithium battery, the second training data is the first feature data of the second-type lithium battery, and i is a positive integer; the multiple above-mentioned training samples are respectively obtained by the first pseudo label. The training samples are input into the model trained in the i-1th training round to obtain multiple output data; based on the multiple training samples, the lithium plating category label of at least one of the first training data and the current pseudo-label of at least one of the second training data, the model parameters of the model trained in the i-1th training round are updated to obtain the model trained in the i-th training round, and when i=1, the current pseudo-label of the second training data is represented as the initial pseudo-label of the second training data, and when the training conditions are met, the model trained in the i-th training round is represented as the lithium plating detection model.

[0008] According to an embodiment of the present invention, the method further includes: when the training condition is not met, updating a current pseudo label of at least one of the second training data based on output data corresponding to each of the at least one of the second training data.

[0009] According to an embodiment of the present invention, the above-mentioned updating of the current pseudo-label of at least one of the second training data based on the output data corresponding to each of the at least one of the second training data includes: determining the maximum value of the predicted probability density distribution corresponding to each of the at least one of the second training data based on the output data corresponding to each of the at least one of the second training data; when the maximum value of the predicted probability density distribution corresponding to the second training data is greater than a preset threshold, updating the current pseudo-label of the second training data based on the output data corresponding to the second training data to obtain an updated pseudo-label; and when the maximum value of the predicted probability density distribution corresponding to the second training data is less than or equal to the preset threshold, determining the updated pseudo-label as the current pseudo-label of the second training data.

[0010] According to an embodiment of the present invention, the above-mentioned method also includes: when the above-mentioned training conditions are not met, obtaining a clustering purity judgment result based on the current pseudo-label of at least one of the above-mentioned second training data; and when the above-mentioned clustering purity judgment result indicates that the clustering effect is poor, updating the current pseudo-label of at least one of the above-mentioned second training data based on the output data corresponding to each of the at least one of the above-mentioned second training data.

[0011] According to an embodiment of the present invention, the clustering purity judgment result is obtained based on the current pseudo-label of at least one of the second training data, including: determining at least one target training data from at least one of the second training data, the target training data corresponding to the first characteristic data of the target lithium battery; determining the lithium deposition category label of at least one of the target lithium batteries by using the lithium deposition diagnosis method; and obtaining the clustering purity judgment result based on the current pseudo-label and lithium deposition category label of at least one of the target lithium batteries.

[0012] According to an embodiment of the present invention, the method further comprises: determining at least one of the target lithium batteries as a first type lithium battery.

[0013] According to an embodiment of the present invention, the above-mentioned feature extraction is performed on the operating data of each of the multiple lithium batteries to obtain the first feature data of each of the multiple lithium batteries, including: using multiple feature extraction models to perform feature extraction based on the operating data of the above lithium batteries to obtain multiple electrochemical features; and obtaining the first feature data of the above lithium batteries based on the multiple electrochemical features.

[0014] Another aspect of the present invention provides a lithium plating detection method based on semi-supervised deep clustering, the method comprising: collecting real-time operating data of the battery; performing feature extraction on the real-time operating data of the battery to obtain feature data of the battery; and inputting the feature data of the battery into a lithium plating detection model to obtain a lithium plating detection result; wherein the lithium plating detection model is trained using any of the above-mentioned training methods for the lithium plating detection model based on semi-supervised deep clustering.

[0015] Another aspect of the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described above.

[0016] Another aspect of the present invention provides a computer-readable storage medium storing computer-executable instructions, which are used to implement the above method when executed.

[0017] Another aspect of the present invention provides a computer program product, the computer program product comprising computer executable instructions, the instructions being used to implement the method as described above when being executed.

[0018] According to an embodiment of the present invention, a semi-supervised lithium plating detection model is obtained by training with the first feature data of each of a plurality of lithium batteries and the lithium plating category labels of a small number of first-category lithium batteries, so as to reduce the dependence on label data and reduce the cost and difficulty of data annotation; in addition, the first feature data of each of the plurality of lithium batteries is used to enable the model to learn more operating data information during the training process, thereby improving the robustness and stability of the lithium plating detection model, so that the model can more accurately identify the lithium plating phenomenon of lithium batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0020] Figure 1 A flow chart of a method for training a lithium plating detection model based on semi-supervised deep clustering according to an embodiment of the present invention is shown.

[0021] Figure 2 A flowchart of a method for training a lithium plating detection model based on semi-supervised deep clustering according to a specific embodiment of the present invention is shown.

[0022] Figure 3 An electron microscope image of a battery according to an embodiment of the present invention is shown.

[0023] Figure 4 A flow chart of a lithium plating detection method based on semi-supervised deep clustering according to an embodiment of the present invention is shown.

[0024] Figure 5 A block diagram of a training device for a lithium plating detection model based on semi-supervised deep clustering according to an embodiment of the present invention is shown.

[0025] Figure 6 A block diagram of a lithium deposition detection device based on semi-supervised deep clustering according to an embodiment of the present invention is shown.

[0026] Figure 7 A block diagram of an electronic device suitable for implementing a training method for a lithium deposition detection model based on semi-supervised deep clustering and a lithium deposition detection method according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0027] Below, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present invention. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of embodiments of the present invention. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of concepts of the present invention.

[0028] The terms used herein are only for describing specific embodiments and are not intended to limit the present invention. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0029] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0030] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0031] In the embodiments of the present invention, the collection, updating, analysis, processing, use, transmission, provision, invention, storage and other aspects of the data involved (for example, including but not limited to user personal information) are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information, network security and national security.

[0032] In the embodiment of the present invention, before obtaining or collecting the user's personal information, the user's authorization or consent is obtained.

[0033] Lithium-ion batteries have been widely used in modern portable electronic devices and electric vehicles due to their significant advantages such as high energy density, long cycle life and no memory effect. However, with the continuous advancement of technology and the increasing complexity of application scenarios, lithium-ion batteries are facing more stringent operating conditions, such as high temperature, low temperature, rapid charging and discharging and other extreme conditions, which greatly test the safety and stability of the battery.

[0034] Lithium deposition is one of the important causes of lithium-ion battery failure. Its root cause is that during the charging process, lithium ions undergo a reduction reaction on the surface of the graphite negative electrode to become lithium metal, and abnormal deposition occurs on the surface. The deposited lithium metal is divided into reversible lithium and irreversible lithium. Reversible lithium can be re-embedded into the negative electrode during the subsequent static process, which is less harmful to the battery and may also occur during mild charge and discharge cycles. Irreversible lithium refers to lithium metal losing electrical connection with the negative electrode and being unable to continue to participate in electrochemical reactions, which not only reduces the number of circulated lithium ions and reduces the capacity of the battery, but the accumulation of irreversible lithium may also form lithium dendrites, piercing the diaphragm and causing internal short circuits in the battery, which in turn causes serious safety problems such as thermal runaway. In addition, lithium deposition also exacerbates the aging process of the battery and shortens the battery life.

[0035] At present, the diagnostic methods for lithium plating failures in lithium-ion batteries mainly include non-destructive testing methods based on voltage data and unsupervised learning methods based on multi-dimensional feature clustering. Among them, the non-destructive testing method is limited by the limited amount of information contained in the voltage data, and it is difficult to fully reflect the internal state changes of the battery, which easily affects the diagnostic accuracy and reliability. The unsupervised learning method captures subtle changes in the battery state through multi-dimensional features. However, due to the lack of clear label information in unsupervised learning, the training and evaluation of the model face great challenges, resulting in low accuracy of the diagnostic results.

[0036] In view of this, the embodiments of the present invention can make full use of the operating data of each battery, overcome the limitations of one-dimensional data, and use the first feature data of each lithium battery and the lithium plating category labels of a small number of first-category lithium batteries to train a semi-supervised lithium plating detection model to reduce dependence on label data and reduce the cost and difficulty of data annotation; in addition, the first feature data of each lithium battery is used to enable the model to learn more operating data information during the training process, thereby improving the robustness and stability of the lithium plating detection model, so that the model can more accurately identify the lithium plating phenomenon of lithium batteries.

[0037] Specifically, an embodiment of the present invention provides a training method for a lithium plating detection model based on semi-supervised deep clustering, including: obtaining operating data of each of a plurality of lithium batteries, the plurality of lithium batteries including a plurality of first-category lithium batteries with lithium plating category labels and a plurality of second-category lithium batteries without labels; performing feature extraction on the operating data of each of the plurality of lithium batteries to obtain first feature data of each of the plurality of lithium batteries; performing semi-supervised training on an initial model based on the first feature data of each of the plurality of lithium batteries and the lithium plating category labels of each of the plurality of first-category lithium batteries to obtain a lithium plating detection model.

[0038] Figure 1 A flow chart of a method for training a lithium plating detection model based on semi-supervised deep clustering according to an embodiment of the present invention is shown.

[0039] like Figure 1 As shown, the training method includes operations S101 to S103.

[0040] In operation S101 , operating data of each of a plurality of lithium batteries is acquired, where the plurality of lithium batteries include a plurality of first-type lithium batteries having lithium plating category labels and a plurality of second-type lithium batteries without labels.

[0041] According to an embodiment of the present invention, the operating data of multiple lithium batteries under different cycle conditions can be obtained through sensor testing, or directly acquired from a data storage platform. The operating data of multiple lithium batteries under different cycle conditions is not limited to data such as the terminal voltage, load current, and surface temperature of each battery cell.

[0042] According to a specific embodiment of the present invention, the cycle conditions of the lithium battery may include a conventional battery performance test and a lithium precipitation induction test. Among them, the conventional battery performance test may include a conventional battery capacity test, a federal city driving condition test, a dynamic stress condition test, etc.; the battery lithium precipitation induction test may include a normal temperature overcharge test, a low temperature charge and discharge test, a rate performance test, etc. For example, the voltage, current, temperature changes and lithium precipitation phenomena of the lithium battery during the overcharge process, such as a sudden drop in voltage and an abnormal increase in temperature, can be obtained through a normal temperature overcharge test.

[0043] According to an embodiment of the present invention, the plurality of lithium batteries used to obtain operation data include a plurality of first-type lithium batteries and a plurality of second-type lithium batteries. The first-type lithium batteries can be used to characterize lithium batteries with lithium ion classification labels. The second-type lithium batteries can be used to characterize lithium batteries without lithium ion classification labels.

[0044] According to an embodiment of the present invention, the lithium deposition category label may include a lithium deposition label and a non-lithium deposition label. The lithium deposition category label may be obtained by taking a small amount of lithium batteries as sample batteries and labeling the sample batteries based on the lithium deposition conditions on the surface of the sample batteries. For example, the selected sample batteries are tested for lithium deposition conditions using microscopic imaging techniques such as scanning electron microscopy (SEM), and each lithium deposition category label may take a value of 0 or 1, where 0 may be used to indicate no lithium deposition label, and 1 may be used to indicate the presence of a lithium deposition label in which dendritic lithium dendrites cover graphite particles.

[0045] In operation S102, feature extraction is performed on the operating data of each of the plurality of lithium batteries to obtain first feature data of each of the plurality of lithium batteries.

[0046] According to an embodiment of the present invention, the first characteristic data can be used to reflect the performance, health status and other properties of the lithium battery, and the first characteristic data of each of the multiple lithium batteries can be extracted using a feature extraction model. Among them, the feature extraction model may include but is not limited to an equivalent circuit model, an incremental capacity curve, coulomb efficiency, a voltage platform, an Arrhenius model, etc. The first characteristic data includes but is not limited to the voltage, current, temperature, charge and discharge rate, capacity decay rate, coulomb efficiency, etc. of the lithium battery.

[0047] In operation S103, based on the first feature data of each of the plurality of lithium batteries and the lithium plating category labels of each of the plurality of first-category lithium batteries, semi-supervised training is performed on the initial model to obtain a lithium plating detection model.

[0048] According to an embodiment of the present invention, a deep learning model may be used as an initial model to classify input feature data, etc. The initial model may include a deep neural network, a convolutional neural network, a recurrent neural network, etc.

[0049] According to an embodiment of the present invention, the initial model is semi-supervisedly trained using the first feature data of each of the plurality of lithium batteries and the lithium plating category labels of each of the plurality of first-type lithium batteries to optimize the model parameters and obtain the trained lithium plating detection model. The first feature data of each of the plurality of lithium batteries includes feature data of a plurality of first-type lithium batteries with lithium plating category labels and feature data of a plurality of second-type lithium batteries without labels.

[0050] According to a specific embodiment of the present invention, during the training process, the loss function may include a supervised part and an unsupervised part, wherein, for operating data with a lithium precipitation category label, a loss function suitable for a classification task may be used to measure the difference between the model prediction and the true label. For unlabeled operating data, a regularization or clustering loss function may be used to produce similar outputs for operating data of the same lithium precipitation category, so as to iteratively train the model and obtain a trained lithium precipitation detection model. Among them, the loss function may include but is not limited to functions such as cross entropy loss, sparse cross entropy loss, and information divergence.

[0051] Based on this, the embodiments of the present invention can make full use of the operating data of each battery, overcome the limitations of one-dimensional data, and use the first feature data of each lithium battery and the lithium plating category labels of a small number of first-category lithium batteries to train a semi-supervised lithium plating detection model to reduce dependence on label data and reduce the cost and difficulty of data annotation; in addition, the first feature data of each lithium battery is used to enable the model to learn more operating data information during the training process, thereby improving the robustness and stability of the lithium plating detection model, so that the model can more accurately identify the lithium plating phenomenon of lithium batteries.

[0052] According to an embodiment of the present invention, in combination with the first feature data of each of the multiple lithium batteries and the lithium plating category labels of each of the multiple first-category lithium batteries, semi-supervised training is performed on the initial model to obtain a lithium plating detection model, including: clustering processing is performed based on the first feature data of each of the multiple second-category lithium batteries to determine the initial pseudo-labels of each of the multiple second-category lithium batteries; and the initial model is trained using the first feature data of each of the multiple first-category lithium batteries with lithium plating category labels and the first feature data of each of the multiple second-category lithium batteries with initial pseudo-labels to obtain a lithium plating detection model.

[0053] According to an embodiment of the present invention, for a plurality of unlabeled second-type lithium batteries, an unsupervised clustering algorithm may be used to perform clustering processing on the first feature data of the second-type lithium batteries to generate initial pseudo labels.

[0054] Specifically, based on a clustering algorithm, according to the feature similarity between the second-class lithium batteries, the first feature data of each of the plurality of unlabeled second-class lithium batteries can be clustered to be divided into one or more clusters, each cluster can represent a different battery state or performance level. Based on the clustering results, an initial pseudo-label is generated for each second-class lithium battery. The clustering algorithm may include but is not limited to the K-means clustering algorithm, the hierarchical clustering algorithm, the density-based spatial clustering of applications with noise (DBSCAN) algorithm, etc.

[0055] For example, the K-means algorithm is used to cluster the first feature data of multiple unlabeled second-type lithium batteries to divide them into two clusters, and the two clusters can represent two battery states of "lithium deposition" and "no lithium deposition". Specifically, the expression of the clustering algorithm is as follows:

[0056] (1);

[0057] Among them, Y u represents the initial pseudo labels of multiple second-category lithium batteries; Z u The feature representation of the first feature data of each of the plurality of second-category lithium batteries is represented, K represents the number of clusters, and KMeans represents a function for performing a K-means clustering operation. The feature representation of the first feature data of each of the plurality of second-category lithium batteries can be a matrix, and each row of the matrix can represent a feature vector of the first feature data of a second-category lithium battery.

[0058] According to a specific embodiment of the present invention, the clustering algorithm iteratively updates the cluster center and assigns the first feature data to the nearest cluster according to the distance from the data point of the first feature data to the cluster center until the cluster center no longer changes significantly or reaches a preset number of iterations. After clustering is completed, each first feature data will be assigned to one of the clusters. We can use the index of this cluster as the initial pseudo-label of the first feature data.

[0059] According to an embodiment of the present invention, the initial model is semi-supervisedly trained by combining the first feature data of each of a plurality of first-type lithium batteries with a lithium plating category label and the first feature data of each of a plurality of second-type lithium batteries with an initial pseudo-label to obtain a lithium plating detection model.

[0060] Specifically, when the initial model adopts a deep neural network, the loss function of the supervised part and the unsupervised part can be established, and the parameters such as the weight and bias of the network can be updated by using an optimizer to obtain a lithium precipitation detection model. The optimizer can include but is not limited to a stochastic gradient descent algorithm, an adaptive moment estimation (Adam) optimizer, an adaptive learning rate (AdaGrad) optimizer, and the like.

[0061] For example, the initial model has multiple hidden layers, where the weighted input and activation function output of each layer can be calculated as follows:

[0062] (2);

[0063] (3);

[0064] Among them, z (l) is the weighted input of layer l, a (l) is the activation function output of layer l, W (l) is the weight matrix of the lth layer, b (l) is the bias vector of the lth layer, f (l) is the activation function of the lth layer. Where l = 1, 2, ..., L, and L represents the number of hidden layers. The activation function may include, but is not limited to, a rectified linear unit (ReLU), a hyperbolic tangent (Tanh), and the like.

[0065] After calculating the weighted input and activation function output of each layer, the cross entropy loss function can be used to establish the loss functions of the supervised part and the unsupervised part respectively. Specifically, the expressions of the loss functions of the supervised part and the unsupervised part are as follows:

[0066] (4);

[0067] (5);

[0068] (6);

[0069] in, is a supervised loss, is the unsupervised loss, is the total loss function, N l is the number of labeled data, N u is the number of unlabeled data, C is the number of categories, where C=2, y i,c is the true label of the i-th data, is the pseudo label of the i-th data, p i,c is the predicted probability of the model, Represents a hyperparameter that can be used to balance the contribution of the two parts of the loss.

[0070] In addition, the gradient descent method can also be used to update the network parameters of the deep neural network, wherein the network parameters may include weights and biases. Specifically, the gradient of the total loss function with respect to each network parameter is calculated, and the weights and biases of the network parameters are updated based on the gradient and the learning rate. The expression for updating the network parameters is as follows:

[0071] (7);

[0072] (8);

[0073] in, represents the network parameters, Represents the learning rate, which can be used to control the step size of network parameter updates. represents the gradient of each network parameter.

[0074] Repeat the above steps until the total loss function converges or the model reaches a preset number of iterations, and a lithium deposition detection model can be obtained. In each iteration, the pseudo labels of the unlabeled data can be recalculated to reflect the current learning state of the model.

[0075] Based on this, the embodiment of the present invention trains the initial model through a small amount of first feature data with a lithium plating category label, so that the model can learn a more effective feature representation in lithium plating detection. At the same time, a large amount of unlabeled data is used to obtain pseudo-labels through cluster analysis to further enrich the training set of the model, so that the model can learn on a wider data distribution, thereby learning more generalized and effective features. Therefore, based on the unsupervised cluster analysis model and the supervised deep learning model, a semi-supervised deep clustering model is formed. By introducing the guidance of labeled data, the clustering process is more directional, which helps to improve the quality of the clustering results, so that the lithium plating detection model can more accurately identify the lithium plating state of the lithium battery, and provide a more reliable basis for subsequent maintenance and decision-making.

[0076] According to an embodiment of the present invention, an initial model is trained using first feature data of each of a plurality of first-type lithium batteries having a lithium plating category label and first feature data of each of a plurality of second-type lithium batteries having an initial pseudo-label to obtain a lithium plating detection model, including: in the i-th training round, a plurality of training samples are determined from the first feature data of each of the plurality of lithium batteries, the plurality of training samples include at least one first training data and at least one second training data, the first training data is the first feature data of the first-type lithium battery, the second training data is the first feature data of the second-type lithium battery, and i is a positive integer; the plurality of training samples are input into the model trained in the i-1-th training round to obtain a plurality of output data; based on the plurality of training samples, the lithium plating category label of each of the at least one first training data and the current pseudo-label of each of the at least one second training data, the model parameters of the model trained in the i-1-th training round are updated to obtain the model trained in the i-th training round, where i=1, the current pseudo-label of the second training data is represented as the initial pseudo-label of the second training data, and when the training conditions are met, the model trained in the i-th training round is represented as the lithium plating detection model.

[0077] According to an embodiment of the present invention, when the total loss function converges or the model reaches a preset number of iterations, the lithium plating detection model can be gradually trained and optimized by updating the initial pseudo labels of each of the plurality of second-category lithium batteries.

[0078] Specifically, in the current training round, multiple training samples are determined from the first feature data of each of the multiple lithium batteries, and the multiple training samples include one or more first feature data of the first type of lithium battery with a lithium ion classification label and the first feature data of the second type of lithium battery with a current pseudo label. Among them, the current pseudo label is used to characterize the pseudo label updated to the current round. The selected multiple training samples are input into the model obtained in the previous training round, and the output data of each training sample is obtained by calculation through each layer of the model. Based on the multiple training samples, the lithium ion classification labels of the first training data and the current pseudo labels of the second training data, a loss function is calculated, such as a total loss function, where the total loss function has supervised loss and unsupervised loss. Then, the gradient of the loss function with respect to the model parameters is calculated by the back propagation algorithm, and the model parameters of the previous training round are updated using the gradient descent method or other optimization algorithms to obtain the model of the current training round. When the model meets the training conditions such as reaching the preset number of iterations, the model of the current training round is used as the lithium ion detection model.

[0079] Among them, the output data can be used to represent the model's predicted probability that each training sample belongs to each category.

[0080] According to an embodiment of the present invention, when the training condition is not satisfied, the current pseudo label of each of the at least one second training data is updated based on the output data corresponding to each of the at least one second training data.

[0081] According to an embodiment of the present invention, when the training conditions are not met, such as when the preset number of iterations is not reached, the current pseudo labels of the second training data can be updated based on the predicted probability values ​​of the model for the second training data belonging to each category.

[0082] Based on this, the embodiments of the present invention continuously update the current pseudo-labels during the training process, and the model gradually learns more effective feature representations and classification decision boundaries, which helps to provide more accurate supervision information in subsequent training rounds, improve the accuracy and generalization ability of lithium plating detection, and thus improve the training effect of the model.

[0083] According to an embodiment of the present invention, based on the output data corresponding to each of the at least one second training data, the current pseudo-label of each of the at least one second training data is updated, including: based on the output data corresponding to each of the at least one second training data, determining the maximum value of the predicted probability density distribution corresponding to each of the at least one second training data; when the maximum value of the predicted probability density distribution corresponding to the second training data is greater than a preset threshold, updating the current pseudo-label of the second training data based on the output data corresponding to the second training data to obtain an updated pseudo-label; and when the maximum value of the predicted probability density distribution corresponding to the second training data is less than or equal to the preset threshold, determining the updated pseudo-label as the current pseudo-label of the second training data.

[0084] According to a specific embodiment of the present invention, for each second training data, the model will output a prediction probability vector, which can be expressed as: z=[z 1 ,z 2 ,...,z K ], where K represents the number of clusters. For example, K=2 can represent two category clusters, namely "lithium precipitation" and "non-lithium precipitation". Then, the predicted probability vector can be converted into a predicted probability density distribution using the exponential normalization (Softmax) function, that is, the predicted probability corresponding to each category cluster. The exponential normalization function ensures that the sum of the predicted probabilities of all category clusters is 1, and each probability value is between 0 and 1. Specifically, the expression of the predicted probability density distribution is as follows:

[0085] (9);

[0086] in, represents the predicted probability density distribution, where k=[k 1 , k 2 ],k 1 Corresponding to the “lithium deposition” state category cluster, k 2 Corresponding to the category cluster representing the "no lithium deposition" state.

[0087] According to a specific embodiment of the present invention, by calculating the maximum value of the predicted probability density distribution of each second training data, it is determined whether the current pseudo label of the second training data needs to be updated.

[0088] For example, a confidence threshold is set as a preset threshold, such as 0.8. The confidence threshold can be used to determine whether the model's prediction of each second training data is reliable enough, thereby deciding whether to update its pseudo-label. The choice of this confidence threshold can be determined based on experience or cross-validation to ensure a balance between maintaining the generalization ability of the model and avoiding overfitting. If the maximum value of the predicted probability density distribution is greater than the confidence threshold, the current pseudo-label is updated to the model predicted category. If the maximum value of the predicted probability density distribution is less than or equal to the confidence threshold, the current pseudo-label remains unchanged.

[0089] According to an embodiment of the present invention, the model is iteratively retrained using the updated pseudo-label and its corresponding second training data, and the first training data with the lithium plating category label until the model meets the training conditions.

[0090] According to an embodiment of the present invention, in order to avoid overfitting of the model due to accumulated errors during the training process, a clustering algorithm (such as K-means) can be used regularly to correct the secondary face data with low confidence. The clustering algorithm can help discover the potential structure in the data, thereby providing more accurate pseudo-labels.

[0091] Based on this, the embodiments of the present invention continuously update pseudo labels, retrain models and perform clustering corrections. The semi-supervised deep clustering model can effectively utilize limited labeled data and a large amount of unlabeled data to train the lithium plating detection model, and gradually learn more accurate and generalized feature representations, thereby improving the accuracy and robustness of lithium plating detection.

[0092] According to an embodiment of the present invention, the training method of the lithium plating detection model based on semi-supervised deep clustering also includes: when the training conditions are not met, based on the current pseudo-label of each of the at least one second training data, obtaining a clustering purity judgment result; and when the clustering purity judgment result indicates that the clustering effect is poor, based on the output data corresponding to each of the at least one second training data, updating the current pseudo-label of each of the at least one second training data.

[0093] According to an embodiment of the present invention, a cluster purity judgment result is obtained based on the current pseudo-label of each of at least one second training data, including: determining at least one target training data from the at least one second training data, the target training data corresponding to the first characteristic data of the target lithium battery; using a lithium plating diagnosis method to determine the lithium plating category label of each of the at least one target lithium battery; and obtaining the cluster purity judgment result based on the current pseudo-label and the lithium plating category label of each of the at least one target lithium battery.

[0094] According to an embodiment of the present invention, when the trained model does not meet the training conditions, the clustering effect can be evaluated using clustering external indicators such as clustering purity, normalized mutual information, confusion matrix, adjusted Rand coefficient, etc.

[0095] Specifically, clustering purity can be used to evaluate the clustering effect. For example, one or more of the second training data with the current pseudo-label is selected as the target training data, wherein the target training data should correspond to the first characteristic data of the target lithium battery, so as to be used to train the model so that it can accurately identify and analyze the lithium plating behavior of the target lithium battery. The target lithium battery is analyzed and diagnosed using a lithium plating diagnosis method such as a scanning electron microscope test to determine its respective lithium plating category label, and the target lithium battery is used as a first-class lithium battery with a lithium plating category label. Based on the current pseudo-label and lithium plating category label of the target lithium battery, the clustering purity discrimination result is calculated. If the clustering purity discrimination result indicates that the clustering effect is poor, that is, the calculated clustering purity is lower than the preset threshold, the current pseudo-labels of the target training data can be updated based on the model's predicted probability that each training sample belongs to each category, until the model meets the training conditions, such as the clustering purity reaches the preset threshold, or the model performance is stable and no longer changes. Among them, the clustering purity can be calculated by the following formula:

[0096] (10);

[0097] Among them, P represents the cluster purity, N represents the number of target lithium batteries, and C i is the i-th cluster, L j is the jth lithium extraction category label. Indicates the number of correctly classified second training data.

[0098] Based on this, an embodiment of the present invention combines semi-supervised learning and deep clustering, and uses a clustering purity evaluation method to update the current pseudo-labels of the second training data with poor clustering effects, so as to gradually reduce the difference between the pseudo-labels and the lithium plating category labels, thereby providing strong support for fault diagnosis and health monitoring of lithium batteries by improving clustering purity and optimizing model performance.

[0099] Reference below Figure 2 , the training method of the lithium plating detection model based on semi-supervised deep clustering is further explained in combination with a specific embodiment.

[0100] Figure 2 A flowchart of a method for training a lithium plating detection model based on semi-supervised deep clustering according to a specific embodiment of the present invention is shown.

[0101] like Figure 2As shown, the training method of the lithium plating detection model based on semi-supervised deep clustering may include operations S201 to S210.

[0102] In operation S201 , first feature data of a plurality of second-type lithium batteries are clustered.

[0103] In operation S202 , first feature data of a plurality of second-type lithium batteries are clustered to generate initial pseudo labels.

[0104] In operation S203, based on the deep neural network, the weighted input and activation function output of each layer are calculated.

[0105] In operation S204, a cross entropy loss function is used to establish loss functions of the supervised part and the unsupervised part respectively.

[0106] In operation S205, it is determined whether the loss function converges or reaches the number of iterations. If yes, operation S206 is performed to update the current pseudo label of the second training data. If no, operation S207 is performed to optimize the deep neural network using the gradient descent method.

[0107] In operation S208, the weights and biases are updated, and operation S202 is re-executed.

[0108] In operation S209, it is determined whether the pseudo label and the model meet the training conditions. If yes, operation S210 is performed to obtain a lithium deposition detection model. If no, operation S202 is performed again to train the model.

[0109] According to an embodiment of the present invention, feature extraction is performed on the operating data of each of the multiple lithium batteries to obtain first feature data of each of the multiple lithium batteries, including: using multiple feature extraction models to perform feature extraction based on the operating data of the lithium batteries to obtain multiple electrochemical features; and obtaining the first feature data of the lithium batteries based on the multiple electrochemical features.

[0110] According to an embodiment of the present invention, in operation S102, since the extracted first feature data of each of the plurality of lithium batteries includes high-dimensional features, these high-dimensional features usually include a large amount of redundant information or noise. Therefore, the first feature data of each of the plurality of lithium batteries can be obtained by performing feature extraction and dimensionality reduction processing on the operating data of each of the plurality of lithium batteries using a feature extraction model.

[0111] According to a specific embodiment of the present invention, the equivalent circuit model, the incremental capacity curve, and the coulomb efficiency can be used to perform feature extraction processing on the operating data of multiple lithium batteries to obtain multiple electrochemical characteristics. For example, the equivalent circuit model can use a first-order RC equivalent circuit model, and its state space equation is as follows:

[0112] (11);

[0113] (12);

[0114] Among them, SOC t+1 , SOC t They represent the percentage of the battery’s current stored power to its total capacity, U ocv,t is the open circuit voltage, U T is the battery terminal voltage, U 1,t , U 1,t+1 is the polarization voltage, I t is the working current, C is the rated capacity of the battery; R s is the ohmic internal resistance, R p is the polarization internal resistance, C p is the polarized capacitance, and t represents the time. In this equivalent circuit model, the parameters to be identified include the ohmic internal resistance R s , polarization internal resistance R p , polarization capacitance C p Through model parameter identification, key electrochemical features reflecting the internal state of lithium batteries can be extracted.

[0115] According to a specific embodiment of the present invention, the incremental capacity curve can transform the first-order phase change process in the battery electrochemical reaction into the peak of the curve. The general expression is as follows:

[0116] (13);

[0117] Among them, Q t+1 , Q t Indicates battery capacity, V t+1 、V t represents voltage, dQ represents capacity differential, and dV represents voltage differential.

[0118] According to a specific embodiment of the present invention, coulombic efficiency is used to measure the efficiency of an energy storage system. In the field of electrochemistry, especially in battery technology, coulombic efficiency is an important performance indicator. The expression is as follows:

[0119] (14);

[0120] Where C represents the coulombic efficiency of the battery, Q 1 Indicates the actual discharge capacity, Q 2 Indicates the actual charging capacity.

[0121] According to a specific embodiment of the present invention, after feature extraction processing is performed on the operating data of each of the multiple lithium batteries, the extracted features can be further reduced in dimension, and the high-dimensional data can be mapped to a lower-dimensional space while retaining the local structure of the data, so as to reduce the complexity of semi-supervised learning training. For example, the principal component analysis method can be used to reduce the feature dimension to obtain the first feature data that can be used for model training. Through feature extraction and dimensionality reduction, the high-dimensional data can be mapped to a lower-dimensional space while retaining the local structure of the data, so as to reduce the complexity of semi-supervised learning training.

[0122] According to a specific embodiment of the present invention, the operating data of multiple lithium batteries under different cycle conditions can be collected by sensors, and the collected operating data can be uploaded to a big data platform to store the operating data. In the process of training a lithium plating detection model based on semi-supervised deep clustering, the operating data of multiple lithium batteries can be extracted from the big data platform, and the operating data of multiple lithium batteries can be labeled and preprocessed.

[0123] According to an embodiment of the present invention, the label classification method of the battery may include physical diagnostic methods such as disassembly and visual observation, scanning electron microscope detection, inductively coupled plasma detection, etc. The label classification method of the battery may also include electrochemical diagnostic methods such as capacity decay method, electrochemical impedance spectroscopy, etc. Among them, label classification refers to selecting a small number of batteries, diagnosing the lithium deposition of the batteries, and labeling these batteries. The diagnostic method is, for example, a method using a scanning electron microscope.

[0124] Figure 3 An electron microscope image of a battery according to an embodiment of the present invention is shown.

[0125] like Figure 3 As shown, in Figure 3 In the battery shown in the middle left picture, the graphite particles are not covered by substances such as lithium, so Figure 3 The battery shown in the middle left picture can be labeled as "non-lithium deposition". Figure 3 In the battery shown in the middle right figure, there are dendritic lithium dendrites on the surface of the graphite particles, and the dendritic lithium dendrites have covered the graphite particles. Figure 3 The battery shown in the middle right picture can be labeled as "lithium plating".

[0126] According to an embodiment of the present invention, data preprocessing includes data cleaning, data filling, and data integration. Data cleaning refers to identifying and correcting or deleting errors, inconsistencies, duplications, and outliers in the data, such as extreme jumps in sensor acquisition signals. Data filling refers to filling missing values ​​in massive operating data, such as filling missing values ​​with the average value of sensor acquisition data. Data integration refers to merging multiple data sources or data sets into a unified data set. For example, the battery terminal voltage, load current, and surface temperature data are integrated into one data set.

[0127] Figure 4 A flow chart of a lithium plating detection method based on semi-supervised deep clustering according to an embodiment of the present invention is shown.

[0128] like Figure 4 As shown, the method includes S401~S403.

[0129] In operation S401 , real-time operation data of a battery is collected.

[0130] In operation S402, feature extraction is performed on real-time operation data of the battery to obtain feature data of the battery.

[0131] In operation S403, the characteristic data of the battery is input into a lithium plating detection model to obtain a lithium plating detection result.

[0132] The lithium deposition detection model is trained using the training method of the lithium deposition detection model based on semi-supervised deep clustering. The embodiment of the present invention uses the real-time collected battery operation data as input and outputs whether the battery is lithium deposited to realize the lithium deposition fault diagnosis of the battery.

[0133] According to an embodiment of the present invention, a lithium deposition detection device based on semi-supervised deep clustering includes a charging and discharging module, a data acquisition module, and a lithium deposition diagnosis module. The charging and discharging module is used to charge and discharge the battery pack to obtain real-time operation data of the battery. The charging and discharging module can use a programmable DC power supply. The data acquisition module is used to collect real-time operation data of the battery cells in the battery pack during operation, including voltage, current, temperature and other data. The data acquisition module can use sensors integrated in the development board of the battery management system (BMS) to collect real-time operation data of the battery. The lithium deposition diagnosis module is used to extract features from the real-time operation data of the battery to obtain feature data of the battery, and input the feature data of the battery into a lithium deposition detection model based on semi-supervised deep clustering to obtain lithium deposition detection results and realize lithium deposition diagnosis of the battery. The lithium deposition diagnosis module can use an STM32 microcontroller, but due to the limited storage and computing resources of embedded devices, it is necessary to compress and optimize the deep learning model. By pruning and quantizing the model, the volume and complexity of the model can be reduced while ensuring the accuracy of the model.

[0134] Figure 5 A block diagram of a training device for a lithium plating detection model based on semi-supervised deep clustering according to an embodiment of the present invention is shown.

[0135] like Figure 5 As shown, the training device of the lithium plating detection model based on semi-supervised deep clustering includes an acquisition module 510, a first feature extraction module 520, and a training module 530.

[0136] The acquisition module 510 is used to acquire the operating data of each of the multiple lithium batteries, where the multiple lithium batteries include multiple first-category lithium batteries with lithium plating category labels and multiple second-category lithium batteries without labels.

[0137] The first feature extraction module 520 is used to extract features from the operating data of each of the multiple lithium batteries to obtain first feature data of each of the multiple lithium batteries.

[0138] The training module 530 is used to perform semi-supervised training on the initial model based on the first feature data of each of the multiple lithium batteries and the lithium plating category labels of each of the multiple first-category lithium batteries to obtain a lithium plating detection model.

[0139] According to an embodiment of the present invention, the training module 530 includes a first determination submodule and a training submodule.

[0140] The first determination submodule is used to perform clustering processing based on the first characteristic data of each of the plurality of second-type lithium batteries to determine the initial pseudo labels of each of the plurality of second-type lithium batteries.

[0141] The training submodule is used to train the initial model using the first feature data of each of the multiple first-type lithium batteries with lithium plating category labels and the first feature data of each of the multiple second-type lithium batteries with initial pseudo labels to obtain a lithium plating detection model.

[0142] According to an embodiment of the present invention, the training submodule includes a first determining unit, a second determining unit, and a first updating unit.

[0143] The first determination unit is used to determine multiple training samples from the first characteristic data of each of the multiple lithium batteries in the i-th training round, the multiple training samples include at least one first training data and at least one second training data, the first training data is the first characteristic data of the first type of lithium battery, the second training data is the first characteristic data of the second type of lithium battery, and i is a positive integer.

[0144] The second determining unit is used to input multiple training samples into the model trained in the (i-1)th training round to obtain multiple output data.

[0145] The first updating unit is used to update the model parameters of the model trained in the i-1th training round based on multiple training samples, the lithium plating category label of at least one first training data and the current pseudo label of at least one second training data, so as to obtain the model trained in the i-th training round. When i=1, the current pseudo label of the second training data is represented as the initial pseudo label of the second training data. When the training conditions are met, the model trained in the i-th training round is represented as the lithium plating detection model.

[0146] According to an embodiment of the present invention, the training device of the lithium plating detection model based on semi-supervised deep clustering also includes an updating module.

[0147] The updating module is used to update the current pseudo label of at least one second training data based on the output data corresponding to the at least one second training data when the training condition is not met.

[0148] According to an embodiment of the present invention, the updating module includes a second determining submodule, a third determining submodule, and a fourth determining submodule.

[0149] The second determination submodule is used to determine the maximum value of the prediction probability density distribution corresponding to each of the at least one second training data based on the output data corresponding to each of the at least one second training data.

[0150] The third determination submodule is used to update the current pseudo label of the second training data based on the output data corresponding to the second training data to obtain an updated pseudo label when the maximum value of the predicted probability density distribution corresponding to the second training data is greater than a preset threshold.

[0151] The fourth determination submodule is used to determine the updated pseudo label as the current pseudo label of the second training data when the maximum value of the predicted probability density distribution corresponding to the second training data is less than or equal to a preset threshold.

[0152] According to an embodiment of the present invention, the training submodule includes a third determining unit and a second updating unit.

[0153] The third determination unit is used to obtain a cluster purity determination result based on the current pseudo label of at least one second training data when the training condition is not met.

[0154] The second updating unit is used to update the current pseudo label of at least one second training data based on the output data corresponding to each of the at least one second training data when the cluster purity determination result indicates that the clustering effect is poor.

[0155] According to an embodiment of the present invention, the second updating unit includes a first determining subunit, a second determining subunit, and a third determining subunit.

[0156] The first determining subunit is used to determine at least one target training data from at least one second training data, where the target training data corresponds to the first characteristic data of the target lithium battery.

[0157] The second determination subunit is used to determine the lithium deposition category label of at least one target lithium battery by using the lithium deposition diagnosis method.

[0158] The third determination subunit is used to obtain a cluster purity determination result based on the current pseudo-label and lithium plating category label of at least one target lithium battery.

[0159] According to an embodiment of the present invention, the second updating unit further includes a fourth determining subunit.

[0160] The fourth determination subunit is used to determine at least one target lithium battery as a first type lithium battery.

[0161] According to the embodiment of the present invention, the first feature extraction module 520 includes a fifth determination submodule and a sixth determination submodule.

[0162] The fifth determination submodule is used to use multiple feature extraction models to perform feature extraction based on the operating data of the lithium battery to obtain multiple electrochemical features.

[0163] The sixth determination submodule is used to obtain first characteristic data of the lithium battery based on multiple electrochemical characteristics.

[0164] Figure 6 A block diagram of a lithium deposition detection device based on semi-supervised deep clustering according to an embodiment of the present invention is shown.

[0165] like Figure 6 As shown, the lithium deposition detection device based on semi-supervised deep clustering includes an acquisition module 610 , a second feature extraction module 620 , and a determination module 630 .

[0166] The acquisition module 610 is used to acquire real-time operating data of the battery.

[0167] The second feature extraction module 620 is used to extract features from the real-time operation data of the battery to obtain feature data of the battery.

[0168] The determination module 630 is used to input the characteristic data of the battery into the lithium plating detection model to obtain the lithium plating detection result.

[0169] According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits, or at least part of the functions of any one of them can be implemented in one module. According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits can be split into multiple modules for implementation. According to the embodiments of the present invention, any one or more of the modules, submodules, units, and subunits can be at least partially implemented as hardware circuits, such as field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems on chips, systems on substrates, systems on packages, application specific integrated circuits (ASICs), or can be implemented by hardware or firmware in any other reasonable way of integrating or packaging the circuit, or by any one of the three implementation methods of software, hardware, and firmware, or by a proper combination of any of them. Alternatively, according to the embodiments of the present invention, one or more of the modules, submodules, units, and subunits can be at least partially implemented as computer program modules, and when the computer program modules are run, the corresponding functions can be executed.

[0170] For example, any multiple of the acquisition module 510, the first feature extraction module 520, and the training module 530 can be combined in one module / unit / sub-unit for implementation, or any one of the modules / units / sub-units can be split into multiple modules / units / sub-units. Alternatively, at least part of the functions of one or more of these modules / units / sub-units can be combined with at least part of the functions of other modules / units / sub-units and implemented in one module / unit / sub-unit. According to an embodiment of the present invention, at least one of the acquisition module 510, the first feature extraction module 520, and the training module 530 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, at least one of the acquisition module 510, the first feature extraction module 520, and the training module 530 may be at least partially implemented as a computer program module, and when the computer program module is executed, the corresponding function may be executed.

[0171] It should be noted that the training device of the lithium deposition detection model based on semi-supervised deep clustering and the lithium deposition detection device based on semi-supervised deep clustering in the embodiments of the present invention correspond to the training method of the lithium deposition detection model based on semi-supervised deep clustering and the lithium deposition detection method based on semi-supervised deep clustering in the embodiments of the present invention. The description of the training device of the lithium deposition detection model based on semi-supervised deep clustering and the lithium deposition detection device based on semi-supervised deep clustering specifically refers to the training method of the lithium deposition detection model based on semi-supervised deep clustering and the lithium deposition detection method based on semi-supervised deep clustering, which will not be repeated here.

[0172] Figure 7 A block diagram of an electronic device suitable for implementing a training method for a lithium deposition detection model based on semi-supervised deep clustering and a lithium deposition detection method according to an embodiment of the present invention is shown. Figure 7 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0173] like Figure 7 As shown, the electronic device according to an embodiment of the present invention includes a processor 701, which can perform various appropriate actions and processes according to the program stored in the read-only memory ROM 702 or the program loaded from the storage part 708 to the random access memory RAM 703. The processor 701 may include, for example, a general-purpose microprocessor (such as a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (for example, an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include an on-board memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0174] In RAM 703, various programs and data required for the operation of the electronic device are stored. The processor 701, ROM 702 and RAM 703 are connected to each other via a bus 704. The processor 701 performs various operations of the method flow according to the embodiment of the present invention by executing the programs in ROM 702 and / or RAM 703. It should be noted that the program can also be stored in one or more memories other than ROM 702 and RAM 703. The processor 701 can also perform various operations of the method flow according to the embodiment of the present invention by executing the programs stored in the one or more memories.

[0175] According to an embodiment of the present invention, the electronic device may further include an input / output (I / O) interface 705, which is also connected to the bus 704. The electronic device may further include one or more of the following components connected to the input / output (I / O) interface 705: an input portion 706 including a keyboard, a mouse, etc.; an output portion 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 708 including a hard disk, etc.; and a communication portion 709 including a network interface card such as a LAN card, a modem, etc. The communication portion 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output (I / O) interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed, so that a computer program read therefrom is installed into the storage portion 708 as needed.

[0176] According to an embodiment of the present invention, the method flow according to an embodiment of the present invention can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 709, and / or installed from the removable medium 711. When the computer program is executed by the processor 701, the above-mentioned functions defined in the system of the embodiment of the present invention are executed. According to an embodiment of the present invention, the system, equipment, device, module, unit, etc. described above can be implemented by a computer program module.

[0177] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiment; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present invention is implemented.

[0178] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium. For example, it may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device.

[0179] For example, according to an embodiment of the present invention, the computer-readable storage medium may include the ROM 702 and / or the RAM 703 described above and / or one or more memories other than the ROM 702 and the RAM 703 .

[0180] An embodiment of the present invention also includes a computer program product, which includes a computer program, which contains program code for executing the method provided by the embodiment of the present invention. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the training method of the lithium plating detection model based on semi-supervised deep clustering provided by the embodiment of the present invention.

[0181] When the computer program is executed by the processor 701, the above functions defined in the system / device of the embodiment of the present invention are executed. According to the embodiment of the present invention, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0182] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 709, and / or installed from the removable medium 711. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0183] According to an embodiment of the present invention, the program code for executing the computer program provided by the embodiment of the present invention can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0184] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram may represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box may also occur in an order different from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions. It can be understood by those skilled in the art that the features recorded in the various embodiments of the present invention can be combined and / or combined in various ways, even if such a combination or combination is not explicitly recorded in the present invention. In particular, without departing from the spirit and teachings of the present invention, the features recorded in the various embodiments of the present invention can be combined and / or combined in various ways. All such combinations and / or combinations fall within the scope of the present invention.

[0185] The embodiments of the present invention are described above. However, these embodiments are only for the purpose of illustration, and are not intended to limit the scope of the present invention. Although each embodiment is described above, it does not mean that the measures in each embodiment cannot be used in combination advantageously. Without departing from the scope of the present invention, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present invention.

Claims

1. A training method for a lithium deposition detection model based on semi-supervised deep clustering, characterized in that: The method comprises: Acquire operating data of each of a plurality of lithium batteries, wherein the plurality of lithium batteries include a plurality of first-category lithium batteries with lithium plating category labels and a plurality of second-category lithium batteries without labels; Extracting features from the operating data of each of the plurality of lithium batteries to obtain first feature data of each of the plurality of lithium batteries; and Based on the first characteristic data of each of the plurality of lithium batteries and the lithium plating category labels of each of the plurality of the first-type lithium batteries, semi-supervised training is performed on the initial model to obtain a lithium plating detection model.

2. The method according to claim 1, characterized in that The method of performing semi-supervised training on the initial model based on the first feature data of each of the plurality of lithium batteries and the lithium deposition category labels of each of the plurality of first-type lithium batteries to obtain a lithium deposition detection model includes: Performing clustering processing based on the first characteristic data of each of the plurality of second-type lithium batteries to determine the initial pseudo labels of each of the plurality of second-type lithium batteries; and The initial model is trained using the first feature data of each of the plurality of first-type lithium batteries having lithium plating category labels and the first feature data of each of the plurality of second-type lithium batteries having initial pseudo labels to obtain the lithium plating detection model.

3. The method according to claim 2, characterized in that The method of using the first feature data of each of the plurality of first-type lithium batteries with lithium deposition category labels and the first feature data of each of the plurality of second-type lithium batteries with initial pseudo labels to train the initial model to obtain the lithium deposition detection model includes: In the i-th training round, a plurality of training samples are determined from the first characteristic data of each of the plurality of lithium batteries, the plurality of training samples include at least one first training data and at least one second training data, the first training data is the first characteristic data of the first type of lithium batteries, the second training data is the first characteristic data of the second type of lithium batteries, and i is a positive integer; Inputting a plurality of the training samples into the model trained in the (i-1)th training round to obtain a plurality of output data; Based on the multiple training samples, the lithium plating category label of at least one of the first training data and the current pseudo label of at least one of the second training data, the model parameters of the model trained in the i-1th training round are updated to obtain the model trained in the i-th training round. When i=1, the current pseudo label of the second training data is represented as the initial pseudo label of the second training data. When the training conditions are met, the model trained in the i-th training round is represented as the lithium plating detection model.

4. The method according to claim 3, characterized in that Also includes: When the training condition is not satisfied, a current pseudo label of at least one of the second training data is updated based on the output data corresponding to the at least one of the second training data.

5. The method according to claim 4, characterized in that The updating of the current pseudo label of at least one of the second training data based on the output data corresponding to the at least one of the second training data comprises: Determine, based on the output data corresponding to at least one of the second training data, a maximum value of the prediction probability density distribution corresponding to at least one of the second training data; When the maximum value of the predicted probability density distribution corresponding to the second training data is greater than a preset threshold, updating the current pseudo label of the second training data based on the output data corresponding to the second training data to obtain an updated pseudo label; and When the maximum value of the predicted probability density distribution corresponding to the second training data is less than or equal to the preset threshold, the updated pseudo label is determined to be the current pseudo label of the second training data.

6. The method according to claim 3, characterized in that Also includes: When the training condition is not met, obtaining a cluster purity determination result based on at least one current pseudo label of each of the second training data; as well as When the cluster purity determination result indicates that the clustering effect is poor, the current pseudo label of at least one of the second training data is updated based on the output data corresponding to the at least one of the second training data.

7. The method according to claim 3, characterized in that The obtaining a cluster purity determination result based on at least one current pseudo label of each of the second training data includes: Determine at least one target training data from at least one of the second training data, wherein the target training data corresponds to the first characteristic data of the target lithium battery; Determining the lithium deposition category label of at least one of the target lithium batteries using a lithium deposition diagnostic method; and The cluster purity discrimination result is obtained based on the current pseudo-label and lithium plating category label of at least one of the target lithium batteries.

8. The method according to claim 7, characterized in that Also includes: At least one of the target lithium batteries is determined to be a first type lithium battery.

9. The method according to claim 1, characterized in that: The extracting features of the operating data of each of the plurality of lithium batteries to obtain first feature data of each of the plurality of lithium batteries comprises: Using a plurality of feature extraction models to extract features based on the operating data of the lithium battery to obtain a plurality of electrochemical features; and Based on the multiple electrochemical characteristics, first characteristic data of the lithium battery is obtained.

10. A lithium deposition detection method based on semi-supervised deep clustering, characterized in that: The method comprises: Collect real-time operating data of the battery; Extracting features from the real-time operating data of the battery to obtain feature data of the battery; and Inputting the characteristic data of the battery into a lithium deposition detection model to obtain a lithium deposition detection result; Wherein, the lithium plating detection model is trained using the training method for a lithium plating detection model based on semi-supervised deep clustering as described in any one of claims 1 to 9.

Citation Information

Patent Citations

  • Semi-supervised learning pseudo label assignment method based on clustering fusion

    CN112418331A

  • Unsupervised pedestrian re-identification method based on pseudo-label self-correction

    CN112507901A

  • Image segmentation method based on semi-supervised learning and related equipment

    CN116363365A

  • Reweighted semi-supervised image classification method based on pseudo labels

    CN116385791A

  • Battery lithium precipitation diagnosis method based on multi-dimensional feature clustering

    CN117907839A

Cited By

  • Customized model construction method for electrolyte coulombic efficiency prediction

    CN119274689A

  • Data verification method and system based on federal semi-supervised learning, terminal and storage medium

    CN122241281A