Hard carbon negative electrode sintering monitoring method and device, computer equipment and storage medium

Abstract

The present disclosure provides a hard carbon negative electrode sintering monitoring method, device, computer equipment and storage medium. The method comprises: acquiring a multi-modal sintering state parameter of a hard carbon negative electrode; performing cross-weighted fusion processing on the multi-modal sintering state parameter to obtain a sintering weighted fusion feature; performing weight heteroscedasticity processing on the sintering weighted fusion feature to obtain a sintering dynamic distribution value; sending a sintering failure signal to a hard carbon negative electrode sintering central control system according to the sintering dynamic distribution value to adjust the sintering mode of the hard carbon negative electrode. After collecting the multi-modal sintering state parameter, the current sintering state of the hard carbon negative electrode is determined, and then the multi-modal sintering state parameter is converted into a corresponding sintering dynamic distribution value, which facilitates the determination of the dynamic distribution of the sintering state of the hard carbon negative electrode. Finally, according to the dynamic distribution of the sintering state, the sintering mode of the hard carbon negative electrode is adjusted, which facilitates the dynamic monitoring and adjustment of the hard carbon sintering process.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of hard carbon negative electrode, and particularly relates to a hard carbon negative electrode sintering monitoring method and device, computer equipment and storage medium. BACKGROUND

[0002] As a key negative electrode material of a new generation of high-energy-density batteries, the specific surface area and pore structure of the sintered product of hard carbon directly affect the first coulomb efficiency, rate performance and cycle stability of the battery. In industrial production, the sintering process controls the microstructure and final performance of the material by adjusting parameters such as temperature curve, atmosphere environment and heating rate. In the high-temperature sintering process, the specific surface area of the material not only affects the electrochemical performance and cycle life, but also significantly determines the preparation cost and product yield.

[0003] The current mainstream specific surface area detection method (such as the BET method) needs to destructively sample and offline analyze the cooled finished product sample, which has the limitations of long detection period, inability to non-destructively monitor the whole batch of materials, lack of real-time performance, etc. This leads to problems such as process adjustment lag, abnormal product identification delay and quality evaluation lag, thereby increasing additional production costs. Although the existing technology attempts to use infrared thermography or machine vision for indirect detection, it is mostly limited to single-mode static analysis and is difficult to capture the dynamic evolution law of the material in the sintering process. SUMMARY

[0004] The purpose of the present disclosure is to overcome the deficiencies in the prior art and provide a hard carbon negative electrode sintering monitoring method and device, computer equipment and storage medium for dynamically monitoring and adjusting the hard carbon sintering process.

[0005] The purpose of the present disclosure is achieved by the following technical solutions:

[0006] A hard carbon negative electrode sintering monitoring method, the method comprising:

[0007] Obtaining a multi-modal sintering state parameter of a hard carbon negative electrode;

[0008] Cross-weight fusion processing the multi-modal sintering state parameter to obtain a sintering weighted fusion feature;

[0009] Weight heteroscedasticity processing the sintering weighted fusion feature to obtain a sintering dynamic distribution value;

[0010] Sending a sintering failure signal to a hard carbon negative electrode sintering central control system according to the sintering dynamic distribution value to adjust the sintering mode of the hard carbon negative electrode.

[0011] In one embodiment, obtaining a multi-modal sintering state parameter of a hard carbon negative electrode comprises: obtaining at least two of a sintering spectrum parameter, a sintering infrared parameter and a sintering production parameter of the hard carbon negative electrode.

[0012] In one of the embodiments, the multi-modal sintering state parameters are cross-weighted and fused to obtain a sintering weighted fusion feature, including: performing low-dimensional extraction on each of the multi-modal sintering state parameters to obtain corresponding modal low-dimensional time sequence features.

[0013] In one of the embodiments, the low-dimensional extraction is further performed on each of the multi-modal sintering state parameters to obtain corresponding modal low-dimensional time sequence features, and then includes: performing bidirectional cross operation on any two modal low-dimensional time sequence features to obtain bidirectional attention flow.

[0014] In one of the embodiments, the bidirectional cross operation is further performed on any two modal low-dimensional time sequence features to obtain bidirectional attention flow, and then includes: performing dynamic weighted fusion operation on the bidirectional attention flow to obtain a sintering weighted fusion feature.

[0015] In one of the embodiments, the sintering weighted fusion feature is subjected to weight heteroscedasticity processing to obtain a sintering dynamic distribution value, including: calculating a sintering fusion mean and a sintering fusion variance of the sintering weighted fusion feature.

[0016] In one of the embodiments, the sintering fusion mean and the sintering fusion variance are calculated, and then includes: obtaining a sintering likelihood loss according to the sintering fusion mean and the sintering fusion variance; detecting whether the sintering likelihood loss is greater than or equal to a preset loss; when the sintering likelihood loss is greater than or equal to the preset loss, sending an iterative optimization signal to a hard carbon negative electrode sintering central control system to update the multi-modal sintering state parameters of the sintering multi-modal model and obtain an updated sintering fusion mean.

[0017] A hard carbon negative electrode sintering monitoring device adopts the hard carbon negative electrode sintering monitoring method of any one of the above embodiments, and the device includes: a multi-modal sintering acquisition module, a sintering data processing module, and a sintering regulation module; the multi-modal sintering acquisition module is used to obtain multi-modal sintering state parameters of a hard carbon negative electrode; the sintering data processing module is used to cross-weight and fuse the multi-modal sintering state parameters to obtain a sintering weighted fusion feature; and the sintering data processing module is used to perform weight heteroscedasticity processing on the sintering weighted fusion feature to obtain a sintering dynamic distribution value; and the sintering regulation module is used to send a sintering loss signal to a hard carbon negative electrode sintering central control system according to the sintering dynamic distribution value to adjust a sintering mode of the hard carbon negative electrode.

[0018] A computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the method of any one of the above embodiments when executing the computer program.

[0019] A computer readable storage medium, having stored thereon a computer program, the computer program being executed by a processor to implement the steps of the method according to any one of the preceding embodiments.

[0020] Compared with the prior art, the present disclosure has at least the following advantages:

[0021] After the multi-modal sintering state parameters are collected, the current sintering state of the hard carbon negative electrode is determined, and then the multi-modal sintering state parameters are converted into corresponding sintering dynamic distribution values, so as to facilitate the determination of the dynamic distribution of the sintering state of the hard carbon negative electrode. Finally, according to the dynamic distribution of the sintering state, the sintering mode of the hard carbon negative electrode is adjusted to feedback adjust the sintering mode of the hard carbon negative electrode, so as to facilitate the dynamic monitoring and adjustment of the hard carbon sintering process. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present disclosure, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0023] Figure 1 A flow chart of the hard carbon negative electrode sintering monitoring method in an embodiment;

[0024] Figure 2 A partial flow chart of the hard carbon negative electrode sintering monitoring method in another embodiment;

[0025] Figure 3 A flow chart of the hard carbon negative electrode sintering monitoring method in another embodiment;

[0026] Figure 4 An internal structure diagram of the computer device in an embodiment. DETAILED DESCRIPTION

[0027] In order to facilitate the understanding of the present disclosure, the present disclosure will be described more fully below with reference to the related drawings. The preferred embodiments of the present disclosure are shown in the drawings. However, the present disclosure can be implemented in many different forms, and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present disclosure more thorough and comprehensive.

[0028] It should be noted that when an element is referred to as being "on" another element, it can be directly on the other element or intervening elements can also be present. When an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements can also be present. The terms "vertical", "horizontal", "left", "right" and similar expressions as used herein are for illustrative purposes only and are not meant to be limiting.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terminology used in the description of the disclosure herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0030] The present disclosure relates to a hard carbon negative electrode sintering monitoring method. In one embodiment, the hard carbon negative electrode sintering monitoring method includes obtaining a multi-modal sintering state parameter of a hard carbon negative electrode; performing cross-weighted fusion processing on the multi-modal sintering state parameter to obtain a sintering weighted fusion feature; performing weight heteroscedasticity processing on the sintering weighted fusion feature to obtain a sintering dynamic distribution value; and sending a sintering failure signal to a hard carbon negative electrode sintering monitoring system according to the sintering dynamic distribution value to adjust the sintering mode of the hard carbon negative electrode. After collecting the multi-modal sintering state parameter, the current sintering state of the hard carbon negative electrode is determined, and then the multi-modal sintering state parameter is converted into a corresponding sintering dynamic distribution value, which facilitates the determination of the dynamic distribution of the sintering state of the hard carbon negative electrode. Finally, according to the dynamic distribution of the sintering state, the sintering mode of the hard carbon negative electrode is adjusted to feedback regulate the sintering mode of the hard carbon negative electrode, which facilitates the dynamic monitoring and adjustment of the hard carbon sintering process.

[0031] Referring to Figure 1 which is a flowchart of a hard carbon negative electrode sintering monitoring method according to an embodiment of the present disclosure. The hard carbon negative electrode sintering monitoring method includes some or all of the following steps.

[0032] S100: Obtain a multi-modal sintering state parameter of a hard carbon negative electrode.

[0033] In this embodiment, the multi-modal sintering state parameter is material sintering state data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameter is multi-modal data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameter corresponds to multiple sintering monitoring modes of the hard carbon negative electrode. By collecting the multi-modal sintering state parameter, the current sintering state of the hard carbon negative electrode can be determined. Specifically, the multi-modal sintering state parameter is multiple sintering induction data of the hard carbon negative electrode powder in the sintering furnace, so as to reflect the real-time sintering condition of the hard carbon negative electrode.

[0034] S200: cross-weighted fusion processing is performed on the multi-modal sintering state parameter to obtain a sintering weighted fusion feature.

[0035] In this embodiment, the multi-modal sintering state parameter is material sintering state data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameter is multi-modal data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameter corresponds to multiple sintering monitoring modes of the hard carbon negative electrode. By collecting the multi-modal sintering state parameter, the current sintering state of the hard carbon negative electrode can be determined. Specifically, the multi-modal sintering state parameter is multiple sintering induction data of the hard carbon negative electrode powder in the sintering furnace, so as to reflect the real-time sintering condition of the hard carbon negative electrode. Through cross-weighted fusion processing of the multi-modal sintering state parameter, the multi-modal sintering state parameter is cross-fused, so that multi-modal data fusion is achieved, which facilitates full use of the complementarity between different modalities, makes up for the missing or noise in a single modality, and improves the overall information completeness. Moreover, integrating different modal information extracted into a stable multi-modal representation, that is, the sintering weighted fusion feature, helps to enhance the semantic expression ability and performance of the model.

[0036] S300: weight heteroscedasticity processing is performed on the sintering weighted fusion feature to obtain a sintering dynamic distribution value.

[0037] In the embodiment, the multi-modal sintering state parameter is material sintering state data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameter is multi-modal data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameter corresponds to multiple sintering monitoring modes of the hard carbon negative electrode. By collecting the multi-modal sintering state parameter, the current sintering state of the hard carbon negative electrode can be determined. Specifically, the multi-modal sintering state parameter is multiple sintering induction data of the hard carbon negative electrode powder in the sintering furnace, so as to reflect the real-time sintering condition of the hard carbon negative electrode. By cross-weighted fusion processing of the multi-modal sintering state parameter, the multi-modal data is fused, so as to fully utilize the complementarity between different modes, make up for the missing or noise in a single mode, and improve the overall information completeness. Moreover, integrating different modal information into a stable multi-modal representation, that is, the sintering weighted fusion feature, helps to enhance the semantic expression ability and performance of the model. The sintering weighted fusion feature is a feature extracted after multi-modal data fusion. In order to improve the robustness of the model, by weight heteroscedasticity processing of the sintering weighted fusion feature, it is converted into a corresponding probability distribution instead of a single point estimate, representing the possible value range of the output under the input condition, so as to reflect the specific surface area of the hard carbon negative electrode material sintering finished product.

[0038] S400: send a sintering failure signal to the hard carbon negative electrode sintering central control system according to the sintering dynamic distribution value, so as to adjust the sintering mode of the hard carbon negative electrode.

[0039] In the embodiment, the multi-modal sintering state parameter is material sintering state data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameter is multi-modal data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameter corresponds to multiple sintering monitoring modes of the hard carbon negative electrode. By collecting the multi-modal sintering state parameter, the current sintering state of the hard carbon negative electrode can be determined. Specifically, the multi-modal sintering state parameter is multiple sintering induction data of the hard carbon negative electrode powder in the sintering furnace, so as to reflect the real-time sintering condition of the hard carbon negative electrode. By cross-weighted fusion processing of the multi-modal sintering state parameter, the multi-modal sintering state parameter is cross-fused, so that the multi-modal data is fused, the complementarity between different modes is fully utilized, the missing or noise in a single mode is made up, and the overall information completeness is improved. Moreover, the different modal information extracted is integrated into a stable multi-modal representation, that is, the sintering weighted fusion feature, which helps to enhance the semantic expression ability and performance of the model. The sintering weighted fusion feature is a feature extracted after multi-modal data fusion. In order to improve the robustness of the model, the sintering weighted fusion feature is converted into a corresponding probability distribution instead of a single point estimate through weight heteroscedasticity processing, representing the possible value range of the output under the input condition, so as to reflect the specific surface area of the hard carbon negative electrode material sintered product. The sintering dynamic distribution value is a distribution probability corresponding to the specific surface area of the hard carbon negative electrode in the sintering process, so as to reflect the specific surface area value range of the hard carbon negative electrode in the sintering process, and facilitate determination of the finished product condition of the hard carbon negative electrode. At this time, the sintering failure signal is sent to the hard carbon negative electrode sintering central control system to control the hard carbon negative electrode material sintering process, so that the sintered hard carbon negative electrode meets the finished product requirements.

[0040] In the above embodiment, after collecting the multi-modal sintering state parameter, the current sintering state of the hard carbon negative electrode is determined, and then the multi-modal sintering state parameter is converted into a corresponding sintering dynamic distribution value, so as to determine the dynamic distribution of the sintering state of the hard carbon negative electrode. Finally, according to the dynamic distribution of the sintering state, the sintering mode of the hard carbon negative electrode is adjusted to feedback adjust the sintering mode of the hard carbon negative electrode, so as to dynamically monitor and adjust the hard carbon sintering process.

[0041] In one of the embodiments, the multi-modal sintering state parameters of the hard carbon negative electrode are acquired, including: acquiring at least two of the sintering spectrum parameters, the sintering infrared parameters and the sintering production parameters of the hard carbon negative electrode. In this embodiment, the multi-modal sintering state parameters are the material sintering state data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameters are the multi-modal data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameters correspond to the multiple sintering monitoring modes of the hard carbon negative electrode. Through the acquisition of the multi-modal sintering state parameters, the current sintering state of the hard carbon negative electrode can be determined. Specifically, the multi-modal sintering state parameters are the multiple sintering sensing data of the hard carbon negative electrode powder in the sintering furnace, so as to reflect the real-time sintering situation of the hard carbon negative electrode. The multi-modal sintering state parameters include the sintering spectrum parameters, the sintering infrared parameters and the sintering production parameters of the hard carbon negative electrode. Specifically, the sintering spectrum parameters are the surface chroma and texture characteristics of the hard carbon negative electrode in the sintering process recorded by the hyperspectral imager, the sintering infrared parameters are the sintering temperature field distribution of the hard carbon negative electrode in the sintering process captured by the infrared thermal imager, and the sintering production parameters are the sintering process flow of the hard carbon negative electrode in the sintering process captured by the PLC recording system. Through the acquisition of the multi-modal data of the hard carbon negative electrode in the sintering process, the subsequent deep collaborative fusion of the three-modal heterogeneous data of the hard carbon negative electrode sintering can be facilitated, and the stability, accuracy and generalization ability of the specific surface area prediction of the hard carbon negative electrode material sintering finished product can be significantly improved, so as to adapt to multiple complex sintering conditions.

[0042] In another embodiment, the initial time and time step of the hyperspectral imager, the infrared thermal imager and the PLC recording system are strictly synchronized, and appropriate acquisition frequency is set to accurately capture the subtle changes in the sintering process and avoid redundant information acquisition. The initial time of the hyperspectral imager, the infrared thermal imager and the PLC recording system is set to , and the acquisition frequency is uniformly set to Hz. The hyperspectral imaging data, the infrared thermal imaging data and the production process parameters are recorded once every t seconds, and the production process parameters include the sintering temperature, the heating rate, the constant temperature uniformity, the atmosphere flow, the pressure and the material density.

[0043] The sintering spectrum parameters captured by the hyperspectral imager are denoted as , the single frame dimension is , C1 is the spectral image channel number of the hyperspectral imager, H1 is the spectral image height of the hyperspectral imager, and W1 is the spectral image width of the hyperspectral imager. The sintering infrared parameters captured by the infrared thermal imager are denoted as , the single frame dimension is , C2 is the spectral image channel number of the infrared thermal imager, H2 is the spectral image height of the infrared thermal imager, and W2 is the spectral image width of the infrared thermal imager. The sintering production parameters recorded by the PLC are , the dimension of production parameters at any moment.

[0044] Further, in order to realize the time step synchronization of multi-modal data, the acquisition frequency of each system is kept consistent. The multi-modal sintering state parameters of the hard carbon negative electrode are obtained, and then include:

[0045] detecting whether the PLC recording frequency of the sintering production parameter matches the preset recording frequency;

[0046] When the PLC recording frequency does not match the preset recording frequency, performing a cubic spline interpolation operation on the sintering production parameter.

[0047] In this embodiment, when the PLC recording frequency deviates, a cubic spline interpolation is used to ensure that the production parameter data is aligned with the spectral / thermal image data and does not destroy the original data distribution characteristics. The input multi-modal data set is represented as , wherein N represents the number of samples, represents the ground truth value of the multi-modal data , and y represents the label set of the multi-modal data.

[0048] ​In one of the embodiments, the multi-modal sintering state parameters are cross-weighted and fused to obtain a sintering weighted fusion feature, including: performing low-dimensional extraction operation on each of the multi-modal sintering state parameters to obtain corresponding modal low-dimensional time sequence features. In this embodiment, the multi-modal sintering state parameters are material sintering state data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameters are multiple modal data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameters correspond to multiple sintering monitoring modes of the hard carbon negative electrode. Through the collection of the multi-modal sintering state parameters, the current sintering state of the hard carbon negative electrode can be determined. Specifically, the multi-modal sintering state parameters are multiple sintering induction data of the hard carbon negative electrode powder in the sintering furnace, so as to reflect the real-time sintering condition of the hard carbon negative electrode. Through the cross-weighted and fused processing of the multi-modal sintering state parameters, the multi-modal data fusion is realized, so as to make full use of the complementarity between different modalities, make up for the missing or noise in a single modality, and improve the overall information completeness. Moreover, integrating the extracted different modal information into a stable multi-modal representation, i.e., the sintering weighted fusion feature, helps to enhance the semantic expression ability and performance of the model. The multi-modal sintering state parameters include sintering spectrum parameters, sintering infrared parameters, and sintering production parameters of the hard carbon negative electrode. Specifically, the sintering spectrum parameters are the surface color and texture features of the hard carbon negative electrode recorded by a hyperspectral imager during the sintering process, the sintering infrared parameters are the sintering temperature field distribution of the hard carbon negative electrode captured by an infrared thermal imager during the sintering process, and the sintering production parameters are the sintering process flow of the hard carbon negative electrode captured by a PLC recording system during the sintering process. Through the collection of the multi-modal data during the sintering process of the hard carbon negative electrode, the subsequent deep collaborative fusion of the three-modal heterogeneous data of the hard carbon negative electrode sintering is facilitated, and the stability, accuracy, and generalization ability of the specific surface area prediction of the hard carbon negative electrode material sintering product are significantly improved, which is suitable for various complex sintering conditions. Through the low-dimensional extraction operation on the multi-modal sintering state parameters, the low-dimensional time sequence features of the sintering spectrum parameters, the sintering infrared parameters, and the sintering production parameters are extracted, specifically,

[0049] the sintering spectrum parameters and the sintering infrared parameters are input into the feature extractor composed of , and then the low-dimensional time sequence features of the sintering spectrum parameters and the sintering infrared parameters can be represented as:

[0050]

[0051]

[0052] wherein, feature extractor parameters of sintering spectrum parameters, feature extractor parameters of sintering infrared parameters, and low-dimensional time sequence features of sintering spectrum parameters and sintering infrared parameters, respectively.

[0053] sintering production parameters are input into a feature extractor composed of low-dimensional time sequence features of sintering production parameters can be expressed as:

[0054]

[0055] wherein, feature extractor parameters of sintering production parameters.

[0056] Further, low-dimensional extraction operations are performed on each of the multi-modal sintering state parameters to obtain corresponding modal low-dimensional time sequence features, and then the method further includes: performing bidirectional cross operation on any two modal low-dimensional time sequence features to obtain bidirectional attention flow. In this embodiment, the multi-modal sintering state parameters are material sintering state data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameters are multiple modal data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameters correspond to multiple sintering monitoring modalities of the hard carbon negative electrode. Through the collection of the multi-modal sintering state parameters, the current sintering state of the hard carbon negative electrode can be determined. Specifically, the multi-modal sintering state parameters are multiple sintering induction data of the hard carbon negative electrode powder in the sintering furnace, so as to reflect the real-time sintering situation of the hard carbon negative electrode. Through the cross weighting fusion processing of the multi-modal sintering state parameters, the multi-modal sintering state parameters are cross fused, so that the multi-modal data fusion is realized, the complementarity between different modalities is fully utilized, the missing or noise existing in a single modality is made up, and the overall information completeness is improved. Moreover, the extracted different modal information is integrated into a stable multi-modal representation, that is, the sintering weighted fusion feature, which is helpful to enhance the semantic expression ability and performance of the model.

[0057] ​The multi-modal sintering state parameters include a sintering spectrum parameter, a sintering infrared parameter and a sintering production parameter of the hard carbon negative electrode. Specifically, the sintering spectrum parameter is that a hyperspectral imager records surface chroma and texture characteristics in the sintering process of the hard carbon negative electrode; the sintering infrared parameter is that an infrared thermal imager captures a sintering temperature field distribution in the sintering process of the hard carbon negative electrode; and the sintering production parameter is that a PLC recording system captures a sintering process flow in the sintering process of the hard carbon negative electrode. Through multi-modal data collection in the sintering process of the hard carbon negative electrode, subsequent three-modal heterogeneous data deep collaborative fusion of the sintering of the hard carbon negative electrode is facilitated, the stability, accuracy and generalization ability of the specific surface area prediction of the sintered product of the hard carbon negative electrode material are significantly improved, and various complex sintering conditions are adapted. Through low-dimensional extraction operation on the multi-modal sintering state parameters, low-dimensional time sequence features are extracted from the sintering spectrum parameter, the sintering infrared parameter and the sintering production parameter.

[0058] After the low-dimensional time sequence features of the modes are extracted, in order to realize cross-modal information interaction and fusion of the above three modal data, a cross attention mechanism is used to dynamically generate a weight vector between different modes, so as to realize weighted fusion and alignment of cross-modal information, as shown in Figure 2 . The extracted low-dimensional time sequence feature set of each mode is denoted as , and the low-dimensional time sequence feature of any mode is denoted as . Through a normalization layer and a linear layer, the query matrix , the key matrix and the value matrix are represented as:

[0059]

[0060]

[0061] wherein , , are projection matrices of the matrix , , . The query matrix , the key matrix and the value matrix of the low-dimensional time sequence feature of another mode are obtained from formula (4) and (5). First, an attention flow from the mode to the mode is constructed, and the attention output is represented as:

[0062] (6)

[0063] wherein, denotes the L2 norm, is a very small positive number. max is used to protect weak features from numerical overflow. By equation (6), the modal attention flow to the modal attention flow to the modal , the two unidirectional attention flows are spliced:

[0064]

[0065] It is input to the gating network to adaptively assign weights to obtain the cross-attention modal weight:

[0066]

[0067] wherein, is the parameter of the gating network. The bidirectional information flow is dynamically fused by adaptive weight assignment, and the bidirectional attention flow output can be expressed as:

[0068]

[0069] Further, the bidirectional cross operation is performed on any two modal low-dimensional time sequence features to obtain a bidirectional attention flow, and then the dynamic weighted fusion operation is performed on the bidirectional attention flow to obtain a sintering weighted fusion feature. In this embodiment, the bidirectional attention flow is the output of the cross-attention mechanism based on any two modal low-dimensional time sequence features, and the modal pair formed by the bidirectional attention flow is only the dynamic fusion between two modalities, and there is a master-slave relationship between different modal pairs. Through the cross-attention mechanism and the dynamic weight adaptive mode, the master-slave relationship between modalities can be dynamically balanced when each pair of modal features is fused. For the fusion between modal pairs, the dynamic weight adaptive strategy is used again, so that the model can dynamically select the key modal pair. The overall model adopts two-level weight control, i.e. and the of the fusion layer, to realize adaptive adjustment of the multi-modal weight.

[0070] The bidirectional attention output set of each modal pair is denoted as , and the dynamic weight distribution within the modal pair is realized by the gating network, for example, the balance between the sintering spectrum parameter and the sintering infrared parameter. For the weight between the cross-modal pairs, the gating network is used again for dynamic adaptive adjustment to obtain the fusion modal weight:

[0071]

[0072] wherein, denotes the weight of the bidirectional attention flow , is the parameter of the second gating network. Then the dynamic weighted fusion feature between each modal pair, i.e., the sintering weighted fusion feature, can be represented as:

[0073]

[0074] In one of the embodiments, the sintering weighted fusion feature is subjected to weight heteroscedastic processing to obtain a sintering dynamic distribution value, including: obtaining a sintering fusion mean and a sintering fusion variance of the sintering weighted fusion feature. In this embodiment, the multi-modal sintering state parameter is the material sintering state data of the hard carbon negative electrode in the sintering process, i.e., the multi-modal sintering state parameter is the multi-modal data of the hard carbon negative electrode in the sintering process, that is, the multi-modal sintering state parameter corresponds to multiple sintering monitoring modalities of the hard carbon negative electrode. Through the collection of the multi-modal sintering state parameter, the current sintering state of the hard carbon negative electrode can be determined. Specifically, the multi-modal sintering state parameter is the multiple sintering induction data of the hard carbon negative electrode powder in the sintering furnace, so as to reflect the real-time sintering situation of the hard carbon negative electrode. Through the cross weighted fusion processing of the multi-modal sintering state parameter, the multi-modal data fusion is realized, so as to make full use of the complementarity between different modalities, make up for the missing or noise in a single modality, and improve the overall information completeness. Moreover, integrating the extracted different modal information into a stable multi-modal representation, i.e., the sintering weighted fusion feature, helps to enhance the semantic expression ability and performance of the model. The sintering weighted fusion feature is a feature extracted after multi-modal data fusion. In order to improve the robustness of the model, the sintering weighted fusion feature is converted into a corresponding probability distribution instead of a single point estimate through weight heteroscedastic processing, representing the possible value range of the output under the input condition, so as to reflect the specific surface area of the hard carbon negative electrode material sintering product. The sintering weighted fusion feature as a representation input condition, by obtaining the corresponding sintering fusion mean and sintering fusion variance, the output value range of the specific surface area of the hard carbon negative electrode material sintering product is embodied by the sintering fusion mean and sintering fusion variance, so as to facilitate the determination of the specific surface area of the hard carbon negative electrode material sintering product. Specifically, the heteroscedastic Gaussian distribution form is adopted, i.e.:

[0075]

[0076] wherein, represents the predicted mean, represents the predicted variance, both of which are learned by the regression prediction head, and the specific process is shown in Figure 3 . The heteroscedastic characteristic allows the model to adaptively adjust the variance to generate a suitable uncertainty range. The regression prediction head takes the sintering weighted fusion feature between each modal pair as input, and outputs the corresponding With :

[0077]

[0078] Further, the sintering fusion mean and the sintering fusion variance of the sintering weighted fusion feature are obtained, and then the method further comprises: obtaining a sintering likelihood loss according to the sintering fusion mean and the sintering fusion variance; detecting whether the sintering likelihood loss is greater than or equal to a preset loss; when the sintering likelihood loss is greater than or equal to the preset loss, sending an iterative optimization signal to a hard carbon negative electrode sintering central control system to update the multi-modal sintering state parameter of the sintering multi-modal model, and obtaining an updated sintering fusion mean. In this embodiment, the sintering fusion mean and the sintering fusion variance serve as the benchmark of the sintering likelihood loss, that is, the sintering fusion mean and the sintering fusion variance obtained by the sintering weighted fusion feature are used to reflect the probability distribution of the specific surface area of the hard carbon negative electrode material sintering finished product, so as to determine the possible value range of the specific surface area of the hard carbon negative electrode material sintering finished product. The sintering likelihood loss is used to represent the internal changes of various uncertainty characterization data, which is convenient for predicting the difference between the specific surface area of the hard carbon negative electrode material sintering finished product and the qualified specific surface area.

[0079] The sintering likelihood loss is obtained by minimizing the negative log-likelihood loss, specifically as follows:

[0080]

[0081] Optimizing the model parameter set The parameter update is iteratively optimized by using an Adam optimizer. The loss significantly improves the reliability of the model by taking into account the internal changes of the prediction uncertainty characterization data. When the hyperspectral imager, infrared thermal imager or PLC recording system fails, The loss can drive the model to learn to predict a larger to compensate for the error input to express its uncertainty about the prediction.

[0082] In this way, according to the deviation between the sintering likelihood loss and the preset loss, the sintering fusion variance is adjusted, specifically, when the sintering likelihood loss is greater than or equal to the preset loss, an iterative optimization signal is sent to the hard carbon negative electrode sintering central control system, at this time the sintering fusion variance will be used as the updated variance to optimize the model parameters, that is, a new sintering fusion mean is generated by feedback iteration, which is convenient for training the prediction model of the output sintering fusion mean and sintering fusion variance, so as to facilitate the prediction and adjustment of the specific surface area of the hard carbon negative electrode material sintering finished product by the sintering fusion mean.

[0083] In another embodiment, when the sintering likelihood loss is greater than or equal to a preset loss, an iterative optimization signal is sent to the hard carbon negative electrode sintering central control system to update the multi-modal sintering state parameters of the sintering multi-modal model and obtain an updated sintering fusion mean, and then the method further comprises:

[0084] Returning and re-executing steps S100 to S300, specifically, re-acquiring three-modal data during the sintering process, ensuring that the initial time and time step of each data acquisition system are strictly synchronized. The synchronized multi-modal data is continuously input into the trained prediction model. The model outputs the sintering fusion mean and sintering fusion variance of the specific surface area of the sintered product in real time through feature extraction, cross-attention mechanism and dynamic weighted fusion. At the same time, based on the real-time prediction value and the dynamic weight returned by the model, the process parameters are adjusted, and the specific steps are as follows:

[0085] Detecting whether the sintering likelihood loss is greater than or equal to a preset loss, and then further comprising: when the sintering likelihood loss is less than the preset loss, an iterative optimization signal is sent to the hard carbon negative electrode sintering central control system to update the multi-modal sintering state parameters of the sintering multi-modal model and obtain an updated sintering fusion mean, and then the method further comprises:

[0086] Obtaining the pre-sintering temperature production inter-modal weight from the cross-attention modal weight through the gating network;

[0087] Detecting whether the pre-sintering temperature production inter-modal weight is greater than a first preset modal weight;

[0088] When the pre-sintering temperature production inter-modal weight is greater than the first preset modal weight, a sintering cooling signal is sent to the hard carbon negative electrode sintering central control system to reduce the sintering heating rate.

[0089] In this embodiment, the pre-sintering temperature production inter-modal weight is the cross-modal inter-weight between the sintering infrared parameters and the sintering production parameters of the hard carbon negative electrode material during the low temperature heating of sintering, i.e. the pre-sintering stage. When the pre-sintering temperature production inter-modal weight is greater than the first preset modal weight, it indicates that the temperature modal weight increases. According to the influence degree of the temperature data, such as the infrared thermal image and the PLC production process parameters, on the heating rate in the modal intra-weight, the heating rate needs to be appropriately reduced at this time. Specifically, if the production process parameter and the temperature field correlation weight are high, the heating rate is further reduced to adjust the amplitude and prolong the residence time of the material in the low temperature interval to optimize the microstructure. Conversely, if the sintering fusion mean is greater than or equal to a preset mean, i.e. the predicted mean of the specific surface area is higher than the set threshold, and / or the pre-sintering temperature production inter-modal weight is less than the first preset modal weight, a sintering heating signal is sent to the hard carbon negative electrode sintering central control system to increase the heating rate to speed up the sintering process.

[0090] Further, the detection of whether the pre-burning temperature production inter-modal weight is greater than the first preset modal weight further includes:

[0091] When the pre-burning temperature production inter-modal weight is equal to the first preset modal weight, the pre-burning spectrum temperature inter-modal weight in the cross-attention modal weight is obtained through the gate network;

[0092] The detection of whether the pre-burning spectrum temperature inter-modal weight is greater than the second preset modal weight;

[0093] When the pre-burning spectrum temperature inter-modal weight is greater than the second preset modal weight, the detection of whether the sintering fusion mean value is less than the preset mean value;

[0094] When the sintering fusion mean value is less than the preset mean value, a pre-burning up-regulation signal is sent to the hard carbon negative electrode sintering central control system to increase the pre-burning temperature lower limit and prolong the pre-burning time.

[0095] In the embodiment, the pre-burning spectrum temperature inter-modal weight is greater than the second preset modal weight, indicating that the surface texture of the hard carbon negative electrode material changes during the pre-burning process. At this time, the predicted texture represented by the specific surface area needs to be detected again. The sintering fusion mean value is less than the preset mean value, indicating that the overall temperature of the pre-burning process is too low at this time, and the carbonization reaction and structure reorganization of the hard carbon negative electrode material are relatively slow. In order to promote the carbonization reaction and structure reorganization, a pre-burning up-regulation signal is sent to the hard carbon negative electrode sintering central control system. On the one hand, the pre-burning temperature lower limit is increased to increase the overall temperature of the pre-burning process, and at the same time, the pre-burning time is prolonged, so that the carbonization reaction and structure reorganization of the hard carbon negative electrode sintering are more complete.

[0096] Further, the detection of whether the sintering fusion mean value is less than the preset mean value further includes:

[0097] When the sintering fusion mean value is greater than or equal to the preset mean value, the high-burning temperature production inter-modal weight in the cross-attention modal weight is obtained through the gate network;

[0098] The detection of whether the high-burning temperature production inter-modal weight is greater than the third preset modal weight;

[0099] When the high-burning temperature production inter-modal weight is greater than the third preset modal weight, the detection of whether the sintering fusion variance is greater than the preset variance;

[0100] When the sintering fusion variance is greater than the preset variance, a high-burning down-regulation signal is sent to the hard carbon negative electrode sintering central control system to reduce the highest temperature of the high-temperature sintering stage.

[0101] In the embodiment, the high-temperature production inter-mode weight is an inter-mode weight between a sintering infrared parameter of the hard carbon negative electrode material and a sintering production parameter in the high-temperature sintering stage. When the high-temperature production inter-mode weight is greater than the third preset mode weight and the sintering fusion variance is greater than the preset variance, it indicates that the temperature field distribution of the hard carbon negative electrode material in the high-temperature sintering stage is abnormal, and the highest temperature needs to be strictly controlled and the temperature fluctuation range needs to be reduced. Specifically, the upper limit of the temperature in the high-temperature sintering stage is adjusted downward to reduce the high-temperature sintering temperature fluctuation, thereby improving the sintering process stability of the hard carbon negative electrode.

[0102] In another embodiment, when the sintering fusion variance is less than or equal to the preset variance, a high-temperature up-regulation signal is sent to the hard carbon negative electrode sintering central control system to increase the highest temperature in the high-temperature sintering stage. At this time, the temperature fluctuation range is appropriately relaxed to optimize the graphitization degree and pore structure of the hard carbon negative electrode.

[0103] In another embodiment, the pre-sintering spectral temperature inter-mode weight is less than or equal to the second preset mode weight, and the high-temperature production inter-mode weight is less than or equal to the third preset mode weight. At this time, the hard carbon sintering process is in a normal state, and there is no need to adjust the internal PLC process parameters, and the current state can be maintained.

[0104] The above various preset variables are set in the database, which is convenient for timely extraction, and different preset variables are placed in different storage units, that is, in different storage stacks. In addition, the pre-sintering temperature production inter-mode weight, the pre-sintering spectral temperature inter-mode weight, and the high-temperature production inter-mode weight can be collected by a corresponding processor, for example, by a central processing chip of a sintering data processing module.

[0105] In one of the embodiments, the disclosure also relates to a hard carbon negative electrode sintering monitoring device, which adopts the hard carbon negative electrode sintering monitoring method of any one of the above-mentioned embodiments. The device comprises a multi-modal sintering acquisition module, a sintering data processing module, and a sintering control module. The multi-modal sintering acquisition module is used to acquire multi-modal sintering state parameters of the hard carbon negative electrode. The sintering data processing module is used to perform cross-weight fusion processing on the multi-modal sintering state parameters to obtain a sintering weighted fusion feature, and to perform weight heteroscedasticity processing on the sintering weighted fusion feature to obtain a sintering dynamic distribution value. The sintering control module is used to send a sintering loss signal to the hard carbon negative electrode sintering central control system according to the sintering dynamic distribution value to adjust the sintering mode of the hard carbon negative electrode.

[0106] In the embodiment, after the multi-modal sintering state parameter acquisition module acquires the multi-modal sintering state parameters, the current sintering state of the hard carbon negative electrode is determined, then the sintering data processing module converts the multi-modal sintering state parameters into corresponding sintering dynamic distribution values, so as to determine the dynamic distribution of the sintering state of the hard carbon negative electrode, and finally the sintering regulation module adjusts the sintering mode of the hard carbon negative electrode according to the dynamic distribution of the sintering state, so as to feedback adjust the sintering mode of the hard carbon negative electrode, and facilitate the dynamic monitoring and adjustment of the hard carbon sintering process.

[0107] In one embodiment, a computer device, which can be a server, is provided, and an internal structure diagram of the computer device can be as shown in Figure 4 The computer device includes a processor, a memory and a network interface connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store multi-modal sintering state parameters, sintering weighted fusion features, sintering dynamic distribution values and sintering failure signals. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement a hard carbon negative electrode sintering monitoring method.

[0108] Those skilled in the art can understand that Figure 4 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0109] In one embodiment, the present application also provides a computer device including a memory and a processor, and the memory stores a computer program. The processor executes the computer program to implement the steps in the above method embodiments.

[0110] In one embodiment, the present application also provides a computer readable storage medium having a computer program stored thereon. The computer program is executed by a processor to implement the steps in the above method embodiments.

[0111] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM can be in a variety of forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0112] The above-described embodiments only express several implementation manners of the present disclosure, which are described in detail and specifically, but cannot be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present disclosure, several modifications and improvements can be made, which are all within the protection scope of the present disclosure. Therefore, the protection scope of the patent of the present disclosure should be subject to the appended claims.

Claims

1. A method for monitoring the sintering of hard carbon anodes, characterized in that, include: Obtain multimodal sintering state parameters of hard carbon anode; The multimodal sintering state parameters are subjected to cross-weighted fusion processing to obtain sintering weighted fusion features; The weighted fusion features of the sintering are subjected to weighted heteroscedasticity processing to obtain the dynamic distribution value of the sintering. Based on the sintering dynamic distribution value, a sintering failure signal is sent to the hard carbon anode sintering control system to adjust the sintering mode of the hard carbon anode. Among them, obtaining the multimodal sintering state parameters of the hard carbon anode includes: Obtain at least two of the following parameters for hard carbon anodes: sintering spectral parameters, sintering infrared parameters, and sintering production parameters; The multimodal sintering state parameters are subjected to cross-weighted fusion processing to obtain sintering weighted fusion features, including: Low-dimensional extraction is performed on the multimodal sintering state parameters of each mode to obtain the corresponding low-dimensional temporal features of the modes. Perform a bidirectional crossover operation on any two modal low-dimensional temporal features to obtain a bidirectional attention stream; A dynamic weighted fusion operation is performed on the bidirectional attention stream to obtain sintering weighted fusion features; The sintering weighted fusion features are subjected to weighted heteroscedasticity processing to obtain the sintering dynamic distribution value, including: Calculate the sintering fusion mean and sintering fusion variance of the sintering weighted fusion characteristics; The sintering likelihood loss is obtained based on the sintering fusion mean and the sintering fusion variance. Detect whether the sintering likelihood loss is greater than or equal to a preset loss; When the sintering likelihood loss is greater than or equal to the preset loss, an iterative optimization signal is sent to the hard carbon anode sintering control system to update the multimodal sintering state parameters of the sintering multimodal model and obtain the updated sintering fusion mean.

2. A hard carbon anode sintering monitoring device, wherein the hard carbon anode sintering monitoring device adopts the hard carbon anode sintering monitoring method as described in claim 1, characterized in that, include: A multi-mode sintering acquisition module is used to acquire multi-mode sintering state parameters of hard carbon anodes. A sintering data processing module is used to perform cross-weighted fusion processing on the multimodal sintering state parameters to obtain sintering weighted fusion features; The weighted fusion features of the sintering are subjected to weighted heteroscedasticity processing to obtain the dynamic distribution value of the sintering. A sintering control module is used to send a sintering failure signal to the hard carbon anode sintering control system according to the sintering dynamic distribution value, so as to adjust the sintering mode of the hard carbon anode.

3. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method of claim 1.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method of claim 1.

Images