Purification method and system for electronic-grade lithium hexafluorophosphate

Through the deep learning neural network model, the correlation relationship between drying temperature and vacuum degree is analyzed, and automated control is achieved, which solves the problems of product decomposition and impurity residue during the purification of electronic grade lithium hexafluorophosphate, and improves the purity and quality of the product.

WO2024239499A1PCT designated stage expired Publication Date: 2024-11-28FUJIAN LONGDE NEW ENERGY CO LTD

Patent Information

Application Number
PCT/CN2023/120020
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-23
Filing Date
2023-09-20
Publication Date
2024-11-28

AI Technical Summary

Technical Problem

In the existing electronic grade lithium hexafluorophosphate purification methods, it is difficult to avoid product decomposition or impurity residues in control of drying temperature and vacuum, which affects the purity and quality of the product.

Method used

Using a neural network model based on deep learning, keyframes are extracted through drying monitoring videos, the contextual implicit relationship between drying temperature and vacuum degree is mined, and automated control is achieved to avoid product decomposition and impurity residues.

Benefits of technology

The purity and quality of electronic grade lithium hexafluorophosphate are improved to ensure the stability of the physical and chemical properties of the product.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2023120020_28112024_PF_FP_ABST
    Figure CN2023120020_28112024_PF_FP_ABST
Patent Text Reader

Abstract

The present application relates to the field of intelligent measurement. Particularly disclosed are a purification method and system for electronic-grade lithium hexafluorophosphate. By means of a deep learning-based neural network model, the present application mines the contextual implicit association relationship between the drying temperature and the vacuum degree during purification, so as to achieve automation control of the drying temperature, thus effectively avoiding product decomposition, loss of crystal water and other problems, and improving the purity and quality of electronic-grade lithium hexafluorophosphate products.
Need to check novelty before this filing date? Find Prior Art

Description

Purification method and system for electronic grade lithium hexafluorophosphate Technical Field

[0001] The present application relates to the field of intelligent detection, and more specifically, to a purification method and system for electronic-grade lithium hexafluorophosphate. Background Art

[0002] Electronic-grade lithium hexafluorophosphate (LiPF6) is widely used in lithium-ion batteries due to its excellent electrochemical properties and is one of the key core materials for manufacturing lithium-ion batteries.

[0003] Driven by high-end market demand, the performance of electronic-grade lithium hexafluorophosphate electrolyte has become a significant factor restricting the industry's development. A lithium battery's cycle life, rate charge-discharge performance, and high- and low-temperature performance are all directly linked to the quality of the electronic-grade lithium hexafluorophosphate electrolyte. Electrolyte impurities can also significantly impact the performance of both the positive and negative electrodes during battery operation. Due to its tendency to decompose with water, electronic-grade lithium hexafluorophosphate (LFP) requires high purity, low free acid content, and low moisture content for use in lithium-ion batteries. Therefore, obtaining high-purity electronic-grade LFP is a pressing challenge. Technical issues

[0004] To address these issues, patent CN 103539168B provides a method for purifying electronic-grade lithium hexafluorophosphate. During the actual purification process, low-temperature vacuum drying directly impacts the purity and quality of the product. However, this step presents the following technical issues: Excessively high or low drying temperatures can cause the product to decompose or lose its water of crystallization, thereby reducing its purity or altering its physical and chemical properties. Excessively low vacuum levels can lead to residual impurities, increasing the degree of product contamination.

[0005] Therefore, an optimized purification method for electronic-grade lithium hexafluorophosphate is desired. Technical Solutions

[0006] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a method and system for purifying electronic-grade lithium hexafluorophosphate, which uses a deep learning-based neural network model to mine the contextual implicit correlation between drying temperature and vacuum degree during the purification process, and thereby realizes automatic control of the drying temperature, thereby effectively avoiding problems such as product decomposition and loss of crystal water, and improving the purity and quality of the electronic-grade lithium hexafluorophosphate product.

[0007] According to one aspect of the present application, a method for purifying electronic-grade lithium hexafluorophosphate is provided, comprising:

[0008] Acquire drying temperature values ​​and vacuum degrees at a plurality of predetermined time points within a predetermined time period, as well as a drying monitoring video of the predetermined time period;

[0009] Extracting the image frames corresponding to the plurality of predetermined time points from the drying monitoring video as a plurality of drying monitoring key frames;

[0010] Passing the plurality of drying monitoring key frames through a convolutional neural network model as a filter to obtain a plurality of drying monitoring semantic understanding feature vectors;

[0011] Passing the plurality of dryness monitoring semantic understanding feature vectors through a transformer-based context encoder to obtain a dryness context semantic understanding feature vector;

[0012] After arranging the drying temperature values ​​and vacuum degrees at the plurality of predetermined time points into a vacuum degree input vector and a drying temperature input vector according to the time dimension, the vacuum degree input vector and the drying temperature input vector are associatedly coded to obtain a parameter collaborative input matrix;

[0013] Passing the parameter collaborative input matrix through a convolutional neural network model as a feature extractor to obtain a parameter collaborative feature matrix;

[0014] Calculating a responsiveness transfer of the dry context semantic understanding feature vector relative to the parameter collaborative feature matrix to obtain a classification feature vector; and

[0015] The classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the drying temperature value at the current time point is appropriate.

[0016] In the above-mentioned method for purifying electronic-grade lithium hexafluorophosphate, the multiple drying monitoring key frames are respectively passed through a convolutional neural network model as a filter to obtain multiple drying monitoring semantic understanding feature vectors, including: using each layer of the convolutional neural network model as a filter to perform the following on the input data in the forward pass of the layer: convolution processing on the input data to obtain a convolution feature map; pooling the convolution feature map based on a local feature matrix to obtain a pooled feature map; and nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the convolutional neural network as a filter is the multiple drying monitoring semantic understanding feature vectors, and the input of the first layer of the convolutional neural network as a filter is the multiple drying monitoring key frames.

[0017] In the above-mentioned method for purifying electronic-grade lithium hexafluorophosphate, the multiple drying monitoring semantic understanding feature vectors are passed through a converter-based context encoder to obtain a drying context semantic understanding feature vector, including: one-dimensionally arranging the multiple drying monitoring semantic understanding feature vectors to obtain a global drying monitoring semantic understanding feature vector; calculating the product between the global drying monitoring semantic understanding feature vector and the transpose vector of each drying monitoring semantic understanding feature vector in the multiple drying monitoring semantic understanding feature vectors to obtain multiple self-attention association matrices; respectively normalizing each self-attention association matrix in the multiple self-attention association matrices to obtain multiple standardized self-attention association matrices; passing each standardized self-attention association matrix in the multiple standardized self-attention association matrices through a Softmax classification function to obtain multiple probability values; respectively weighting each drying monitoring semantic understanding feature vector in the multiple drying monitoring semantic understanding feature vectors with each probability value in the multiple probability values ​​as a weight to obtain the multiple contextual semantic drying monitoring semantic understanding feature vectors; and cascading the multiple contextual semantic drying monitoring semantic understanding feature vectors to obtain the drying context semantic understanding feature vector.

[0018] In the above-mentioned method for purifying electronic-grade lithium hexafluorophosphate, after arranging the drying temperature values ​​and vacuum degrees at the plurality of predetermined time points into a vacuum degree input vector and a drying temperature input vector according to the time dimension, the vacuum degree input vector and the drying temperature input vector are associated and encoded to obtain a parameter synergistic input matrix, including: associating and encoding the vacuum degree input vector and the drying temperature input vector using the following association formula to obtain a parameter synergistic input matrix; wherein the formula is: in represents the vacuum degree input vector, represents the transposed vector of the vacuum degree input vector, represents the drying temperature input vector, represents the parameter collaborative input matrix, Represents vector multiplication.

[0019] In the above-mentioned method for purifying electronic-grade lithium hexafluorophosphate, calculating the responsiveness transfer of the dry context semantic understanding feature vector relative to the parameter synergistic feature matrix to obtain a classification feature vector includes: calculating the responsiveness transfer of the dry context semantic understanding feature vector relative to the parameter synergistic feature matrix to obtain a classification feature vector using the following formula; wherein the formula is: in represents the DRY context semantic understanding feature vector, represents the parameter collaborative feature matrix, represents the classification feature vector.

[0020] The above-mentioned method for purifying electronic-grade lithium hexafluorophosphate also includes a training step: training the convolutional neural network model as a filter, the converter-based context encoder, the convolutional neural network model as a feature extractor, and the classifier; wherein the training step includes: obtaining training data, the training data including training drying temperature values ​​and training vacuum degrees at multiple predetermined time points within a predetermined time period, training drying monitoring videos of the predetermined time period, and a true value of whether the drying temperature value at the current time point is appropriate; extracting training image frames corresponding to the multiple predetermined time points from the training drying monitoring video as multiple training drying monitoring key frames; passing the multiple training drying monitoring key frames through the convolutional neural network model as a filter to obtain multiple training drying monitoring semantic understanding feature vectors; passing the multiple training drying monitoring semantic understanding feature vectors through the converter-based context encoder to obtain training drying context semantic understanding feature vectors; passing the multiple predetermined time points After the training drying temperature values ​​and training vacuum degrees are arranged as training vacuum degree input vectors and training drying temperature input vectors according to the time dimension respectively, the training vacuum degree input vectors and training drying temperature input vectors are associatedly encoded to obtain a training parameter collaborative input matrix; the training parameter collaborative input matrix is ​​passed through the convolutional neural network model as a feature extractor to obtain a training parameter collaborative feature matrix; the responsiveness transfer of the training drying context semantic understanding feature vector relative to the training parameter collaborative feature matrix is ​​calculated to obtain a training classification feature vector; the training classification feature vector is periodically reparameterized by the Gumbel normal to obtain a training optimized classification feature vector; the training optimized classification feature vector is passed through a classifier to obtain a classification loss function value; and, based on the classification loss function value and propagated in the direction of gradient descent, the convolutional neural network model as a filter, the converter-based context encoder, the convolutional neural network model as a feature extractor, and the classifier are trained.

[0021] In the above-mentioned method for purifying electronic-grade lithium hexafluorophosphate, performing Gumbel normal periodic reparameterization on the training classification feature vector to obtain a training optimized classification feature vector comprises: performing Gumbel normal periodic reparameterization on the training classification feature vector to obtain a training optimized classification feature vector according to the following formula; wherein the formula is: in, represents the training classification feature vector, represents the training optimized classification feature vector, represents the eigenvalues ​​of each position of the training classification feature vector, and are the mean and variance of the eigenvalue set at each position of the training classification feature vector, and , represents the logarithmic function value with base 2, represents the value of the inverse sine function, represents the value of the arccosine function, Represents the feature value of each position of the optimized training classification feature vector.

[0022] According to another aspect of the present application, a purification system for electronic-grade lithium hexafluorophosphate is provided, comprising:

[0023] An information acquisition module, configured to obtain drying temperature values ​​and vacuum degrees at a plurality of predetermined time points within a predetermined time period, as well as a drying monitoring video of the predetermined time period;

[0024] a key frame extraction module, configured to extract image frames corresponding to the plurality of predetermined time points from the drying monitoring video as a plurality of drying monitoring key frames;

[0025] a first convolution module, configured to pass the plurality of drying monitoring key frames through a convolutional neural network model as a filter to obtain a plurality of drying monitoring semantic understanding feature vectors;

[0026] a context encoding module, configured to pass the plurality of drying monitoring semantic understanding feature vectors through a transformer-based context encoder to obtain a drying context semantic understanding feature vector;

[0027] an association coding module, configured to arrange the drying temperature values ​​and vacuum degrees at the plurality of predetermined time points into a vacuum degree input vector and a drying temperature input vector according to the time dimension, and then perform association coding on the vacuum degree input vector and the drying temperature input vector to obtain a parameter coordination input matrix;

[0028] A second convolution module is configured to pass the parameter collaborative input matrix through a convolutional neural network model as a feature extractor to obtain a parameter collaborative feature matrix;

[0029] a responsiveness transfer module, configured to calculate a responsiveness transfer of the dry context semantic understanding feature vector relative to the parameter collaborative feature matrix to obtain a classification feature vector; and

[0030] The classification result generating module is used to pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the drying temperature value at the current time point is appropriate.

[0031] According to another aspect of the present application, an electronic device is provided, comprising: a processor; and a memory, wherein computer program instructions are stored in the memory, and when the computer program instructions are executed by the processor, the processor executes the method for purifying electronic-grade lithium hexafluorophosphate as described above.

[0032] According to another aspect of the present application, a computer-readable medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the method for purifying electronic-grade lithium hexafluorophosphate as described above. Beneficial effects

[0033] Compared with the prior art, the present application provides a method and system for purifying electronic-grade lithium hexafluorophosphate, which uses a deep learning-based neural network model to mine the contextual implicit correlation between drying temperature and vacuum degree during the purification process, and thereby realizes automatic control of the drying temperature, thereby effectively avoiding problems such as product decomposition and loss of crystal water, and improving the purity and quality of the electronic-grade lithium hexafluorophosphate product. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0035] FIG1 is a schematic diagram of a method for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application;

[0036] FIG2 is a flow chart of the inference stage in the method for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application;

[0037] FIG3 is a flow chart of a training phase in a method for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application;

[0038] FIG4 is a schematic diagram of the architecture of the inference stage in the method for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application;

[0039] FIG5 is a schematic diagram of the architecture of the training phase in the method for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application;

[0040] FIG6 is a flow chart of convolutional neural network encoding in the method for purifying electronic grade lithium hexafluorophosphate according to an embodiment of the present application;

[0041] FIG7 is a flowchart of a context encoding process in a method for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application;

[0042] FIG8 is a block diagram of a purification system for electronic-grade lithium hexafluorophosphate according to an embodiment of the present application;

[0043] FIG9 is a block diagram of an electronic device according to an embodiment of the present application. Modes for Carrying Out the Invention

[0044] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0045] Scenario Overview: In response to the above technical problems, the technical concept of this application is: to use deep learning and artificial intelligence technology to extract high-level implicit correlation feature information from the drying temperature values ​​and vacuum degrees at multiple predetermined time points and the drying monitoring videos of a predetermined time period, and to realize automatic control of the drying temperature through classification processing, thereby effectively avoiding problems such as product decomposition and loss of crystal water, and improving the purity and quality of electronic-grade lithium hexafluorophosphate products.

[0046] Specifically, in the technical solution of the present application, first, the drying temperature values ​​and vacuum levels at multiple predetermined time points within a predetermined time period are obtained, as well as a drying monitoring video for the predetermined time period. It should be understood that the drying temperature values ​​and vacuum levels at the multiple predetermined time points and the drying monitoring video for the predetermined time period can comprehensively monitor the drying process. During implementation, the drying temperature values ​​and vacuum levels at the multiple predetermined time points can be monitored in real time using thermometers and vacuum gauges installed on the drying equipment. At the same time, cameras or other monitoring equipment are installed near the drying equipment to record the drying process in real time.

[0047] Next, image frames corresponding to the predetermined time points are extracted from the drying monitoring video as multiple drying monitoring key frames. By segmenting the drying monitoring video into a series of key frames, the changes and detailed features of the drying process can be better captured. Compared to recording and analyzing the entire drying monitoring video, extracting only the key frames significantly reduces data volume and computational complexity, retaining useful drying parameter data while eliminating unnecessary noise interference. Therefore, extracting the required drying monitoring key frames from the drying monitoring video is an efficient and accurate data processing method.

[0048] Then, the multiple drying monitoring key frames are respectively passed through a convolutional neural network model as a filter to obtain multiple drying monitoring semantic understanding feature vectors. Here, by using a convolutional neural network model, efficient and accurate image feature extraction can be performed on the drying monitoring key frames, thereby better capturing the detailed changes and regularity characteristics of the drying process, which is very important for the subsequent analysis and calculation of drying parameter data. Among them, the convolutional neural network model as a filter is a deep learning model used to process two-dimensional data such as images. It is based on convolution operations and pooling operations and has excellent performance in feature extraction. The model is usually composed of multiple convolution layers and pooling layers, wherein the convolution layer can extract and transform the input data through the convolution kernel, while the pooling layer is used to downsample the feature map. When extracting features for the drying monitoring key frames, the convolutional neural network model as a filter performs convolution calculations on each key frame and transfers information between network layers. The model can automatically learn and extract complex feature information in the key frames and represent it as a semantic understanding feature vector.

[0049] In order to better capture the information and global correlation in multiple drying monitoring key frames, in the technical solution of the present application, the multiple drying monitoring semantic understanding feature vectors are passed through a transformer-based context encoder to obtain a drying context semantic understanding feature vector. Here, the transformer-based context encoder (Transformer-based Contextual Encoder) is a deep learning model for processing sequence data and time series data. It is mainly based on the self-attention mechanism and can adaptively assign different weights to each element in the input sequence, so as to better capture the important information in the sequence. This model has good performance and efficiency when processing long sequence or time series data, and is widely used in natural language processing, image processing and other fields.

[0050] During the drying process, there is an implicit collaborative correlation between the drying temperature and the vacuum degree. In order to explore this correlation, in the technical solution of the present application, first, the drying temperature values ​​and vacuum degrees at the multiple predetermined time points are arranged according to the time dimension as a vacuum degree input vector and a drying temperature input vector, respectively, so as to construct the time series discrete distribution of the drying temperature values ​​and the time series discrete distribution of the vacuum degree into structured vacuum degree input vectors and drying temperature input vectors. Then, the vacuum degree input vector and the drying temperature input vector are associatedly encoded to characterize the correlation between the vacuum degree and the drying temperature values ​​at different time points, thereby obtaining a parameter collaborative input matrix. Then, the parameter collaborative input matrix is ​​passed through a convolutional neural network model as a feature extractor to extract feature information related to the collaborative changes of the drying parameters, thereby obtaining a parameter collaborative feature matrix. The parameter collaborative feature matrix contains the complex interactive relationship between different parameters in the drying process and has a strong expressive ability.

[0051] Then, the responsiveness transfer of the drying context semantic understanding feature vector relative to the parameter synergy feature matrix is ​​calculated to obtain a classification feature vector. Here, the parameter synergy feature matrix reflects the complex mutual influence and synergy effects contained between different drying parameters, while the drying context semantic understanding feature vector can reflect the state of the entire drying process. In the technical solution of this application, responsiveness transfer is used to characterize the responsive logical relationship between the two to optimize the effect of subsequent classification.

[0052] After obtaining the classification feature vector, it is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the drying temperature value at the current time point is appropriate. Here, the classifier is a binary classifier, which can learn the relationship between the classification feature vector and the actual drying temperature value. Specifically, in actual applications, if the classification result is "appropriate", it means that the drying temperature value at the current time point meets the requirements; if the classification result is "unsuitable", it means that the drying temperature value at the current time point has a deviation or abnormality, and needs to be adjusted and controlled. In this way, problems such as product decomposition and loss of crystal water caused by excessively high or low drying temperatures are effectively avoided, thereby improving the purity and quality of electronic-grade lithium hexafluorophosphate products.

[0053] In the technical solution of the present application, since the classification feature vector is obtained by calculating the responsive transfer of the dry context semantic understanding feature vector relative to the parameter collaborative feature matrix, the feature distribution of the classification feature vector essentially represents the domain transfer feature between the monitoring frame context image semantics expressed by the dry context semantic understanding feature vector and the correlation feature of the drying temperature value and vacuum degree expressed by the parameter collaborative feature matrix. Therefore, due to the difference between the feature domains of the dry context semantic understanding feature vector and the parameter collaborative feature matrix itself, it is easy for a distribution gap to exist in the distribution of the domain transfer feature, which will lead to poor continuity of the overall feature distribution of the classification feature vector, affecting the training effect during model training.

[0054] Based on this, the applicant of this application has Perform Gumbel normal periodic reparameterization to obtain the optimized classification feature vector , specifically expressed as: , and The eigenvalue sets are The mean and variance of .

[0055] Here, the Gumbel normal periodicity is reparameterized by replacing the categorical eigenvectors The eigenvalues ​​at each position Convert it into the angular feature expression of its probability distribution, and introduce random periodic distribution into the normal distribution of the eigenvalue set based on the random periodic operation of Gumbel distribution, so as to obtain a random periodic continuous differentiable approximation of the original feature distribution, thereby improving the optimized classification feature vector through periodic reparameterization of the feature. During training, the gradient of the loss function is dynamically and continuously fluctuated in the back propagation of the model to improve the dynamic applicability of the convolution kernel during the training process, thereby compensating for the impact of the poor continuity of the feature distribution of the classification feature vector on the training effect, such as training speed and the accuracy of the convergence result.

[0056] Based on this, the present application proposes a method for purifying electronic-grade lithium hexafluorophosphate, which includes: obtaining drying temperature values ​​and vacuum degrees at multiple predetermined time points within a predetermined time period, as well as a drying monitoring video of the predetermined time period; extracting image frames corresponding to the multiple predetermined time points from the drying monitoring video as multiple drying monitoring key frames; passing the multiple drying monitoring key frames through a convolutional neural network model as a filter to obtain multiple drying monitoring semantic understanding feature vectors; passing the multiple drying monitoring semantic understanding feature vectors through a converter-based context encoder to obtain a drying context semantic understanding feature vector; arranging the drying temperature values ​​and vacuum degrees at the multiple predetermined time points into a vacuum input vector and a drying temperature input vector according to the time dimension, and then associatively encoding the vacuum input vector and the drying temperature input vector to obtain a parameter synergistic input matrix; passing the parameter synergistic input matrix through a convolutional neural network model as a feature extractor to obtain a parameter synergistic feature matrix; calculating the responsiveness transfer of the drying context semantic understanding feature vector relative to the parameter synergistic feature matrix to obtain a classification feature vector; and passing the classification feature vector through a classifier to obtain a classification result, which is used to indicate whether the drying temperature value at the current time point is appropriate.

[0057] Figure 1 is a schematic diagram of a scenario for a method for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application. As shown in Figure 1 , in this application scenario, a thermometer (e.g., T, as shown in Figure 1 ) is used to obtain drying temperature values ​​at multiple predetermined time points within a predetermined time period. A vacuum gauge (e.g., V, as shown in Figure 1 ) is used to obtain vacuum levels at multiple predetermined time points within a predetermined time period. Furthermore, a camera (e.g., C, as shown in Figure 1 ) is used to capture drying monitoring video for the predetermined time period. This information is then input into a server (e.g., cloud server S, as shown in Figure 1 ) that deploys an algorithm for purifying electronic-grade lithium hexafluorophosphate. The server processes this input information using the algorithm for purifying electronic-grade lithium hexafluorophosphate to generate a classification result indicating whether the drying temperature value at the current time point is appropriate.

[0058] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0059] Exemplary method: Figure 2 is a flow chart of the inference stage in the method for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application. As shown in Figure 2, the method for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application includes: the inference stage, including: S110, obtaining the drying temperature value and vacuum degree at multiple predetermined time points within a predetermined time period, and the drying monitoring video of the predetermined time period; S120, extracting the image frames corresponding to the multiple predetermined time points from the drying monitoring video as multiple drying monitoring key frames; S130, passing the multiple drying monitoring key frames through a convolutional neural network model as a filter to obtain multiple drying monitoring semantic understanding feature vectors; S140, passing the multiple drying monitoring semantic understanding feature vectors through a converter-based context encoder to obtain a drying context semantic understanding feature vector; S15 0, after arranging the drying temperature values ​​and vacuum degrees at the multiple predetermined time points into a vacuum degree input vector and a drying temperature input vector according to the time dimension respectively, the vacuum degree input vector and the drying temperature input vector are associatedly encoded to obtain a parameter collaborative input matrix; S160, the parameter collaborative input matrix is ​​passed through a convolutional neural network model as a feature extractor to obtain a parameter collaborative feature matrix; S170, the responsiveness transfer of the drying context semantic understanding feature vector relative to the parameter collaborative feature matrix is ​​calculated to obtain a classification feature vector; and, S180, the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the drying temperature value at the current time point is appropriate.

[0060] FIG4 is a schematic diagram of the architecture of the inference stage in the method for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application. As shown in Figure 4, in the inference stage, in this network structure, first, the drying temperature values ​​and vacuum levels at multiple predetermined time points within a predetermined time period, as well as the drying monitoring video of the predetermined time period, are obtained; secondly, image frames corresponding to the multiple predetermined time points are extracted from the drying monitoring video as multiple drying monitoring key frames; the multiple drying monitoring key frames are then respectively passed through a convolutional neural network model as a filter to obtain multiple drying monitoring semantic understanding feature vectors; the multiple drying monitoring semantic understanding feature vectors are passed through a converter-based context encoder to obtain a drying context semantic understanding feature vector; then, the drying temperature values ​​and vacuum levels at the multiple predetermined time points are arranged according to the time dimension as a vacuum input vector and a drying temperature input vector, and then the vacuum input vector and the drying temperature input vector are associatively encoded to obtain a parameter synergistic input matrix; then, the parameter synergistic input matrix is ​​passed through a convolutional neural network model as a feature extractor to obtain a parameter synergistic feature matrix; the responsiveness transfer of the drying context semantic understanding feature vector relative to the parameter synergistic feature matrix is ​​calculated to obtain a classification feature vector; and then, the classification feature vector is passed through a classifier to obtain a classification result, which is used to indicate whether the drying temperature value at the current time point is appropriate.

[0061] Specifically, in step S110, the drying temperature and vacuum level at multiple predetermined time points within a predetermined time period are obtained, as well as a drying monitoring video for the predetermined time period. It should be understood that in the actual purification process, the low-temperature vacuum drying step directly affects the purity and quality of the product. In one example, too high or too low a drying temperature may cause the product to decompose or lose water of crystallization, thereby reducing the purity of the product or changing its physicochemical properties, while too low a vacuum level may result in residual impurities and increase the degree of product contamination. Therefore, in order to effectively avoid problems such as product decomposition and loss of water of crystallization, and improve the purity and quality of the electronic-grade lithium hexafluorophosphate product, the technical solution of the present application analyzes the drying temperature and vacuum level during the purification process, and based on this, realizes automated control of the drying temperature. Therefore, in a specific example of the present application, first, the drying temperature values ​​at multiple predetermined time points within the predetermined time period can be obtained using a thermometer, the vacuum level can be obtained using a vacuum gauge, and the drying monitoring video for the predetermined time period can be obtained using a camera.

[0062] Specifically, in step S120, the image frames corresponding to the multiple predetermined time points are extracted from the drying monitoring video as multiple drying monitoring key frames. In the technical solution of the present application, the image frames corresponding to the multiple predetermined time points are extracted from the drying monitoring video as multiple drying monitoring key frames. By dividing the drying monitoring video into a series of key frames, the changes and detailed features of the drying process can be better captured. Compared with recording and analyzing the drying monitoring video throughout the process, extracting only the key frames can greatly reduce the amount of data and computational complexity, and can retain useful drying parameter data while eliminating unnecessary noise interference. Therefore, extracting the required drying monitoring key frames from the drying monitoring video is an efficient and accurate data processing method.

[0063] Specifically, in step S130, the multiple drying monitoring key frames are respectively passed through a convolutional neural network model as a filter to obtain multiple drying monitoring semantic understanding feature vectors. The multiple drying monitoring key frames are respectively passed through the convolutional neural network model as a filter to obtain multiple drying monitoring semantic understanding feature vectors. Using the convolutional neural network model, efficient and accurate image feature extraction can be performed on the drying monitoring key frames, thereby better capturing detailed changes and regularity characteristics during the drying process, which is very important for the subsequent analysis and calculation of drying parameter data. The convolutional neural network model serving as a filter is a deep learning model used to process two-dimensional data such as images. In one specific example, the convolutional neural network includes multiple cascaded neural network layers, each of which includes a convolution layer, a pooling layer, and an activation layer. During the encoding process of the convolutional neural network, each layer of the convolutional neural network performs convolution processing on the input data using a convolution kernel, pools the convolution feature map output by the convolution layer using the pooling layer, and activates the pooled feature map output by the pooling layer using the activation layer during the forward pass of the layer. When extracting features from drying monitoring key frames, the convolutional neural network model acting as a filter performs convolution calculations on each key frame and transfers information between network layers. The model can automatically learn and extract complex feature information from key frames and represent it as a semantic understanding feature vector.

[0064] FIG6 is a flow chart of convolutional neural network encoding in the method for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application. As shown in FIG6 , the encoding process of the convolutional neural network includes: using each layer of the convolutional neural network model as a filter to perform the following steps on the input data in the forward pass of the layer: S210, convolution processing on the input data to obtain a convolution feature map; S220, pooling the convolution feature map based on a local feature matrix to obtain a pooled feature map; and, S230, performing nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the convolutional neural network as a filter is the multiple drying monitoring semantic understanding feature vectors, and the input of the first layer of the convolutional neural network as a filter is the multiple drying monitoring key frames.

[0065] Specifically, in step S140, the multiple dry monitoring semantic understanding feature vectors are passed through a transformer-based context encoder to obtain a dry context semantic understanding feature vector. That is, the multiple dry monitoring semantic understanding feature vectors are passed through a transformer-based context encoder to obtain a dry context semantic understanding feature vector. Here, the transformer-based context encoder (Transformer-based Contextual Encoder) is a deep learning model used to process sequence data and time series data. It is mainly based on the self-attention mechanism and can adaptively assign different weights to each element in the input sequence, so as to better capture the important information in the sequence. This model has good performance and efficiency when processing long sequence or time series data, and is widely used in natural language processing, image processing and other fields.

[0066] FIG7 is a flowchart of a context encoding process in a method for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application. As shown in Figure 7, the context encoding process includes: S310, arranging the multiple dry monitoring semantic understanding feature vectors in one dimension to obtain a global dry monitoring semantic understanding feature vector; S320, calculating the product between the global dry monitoring semantic understanding feature vector and the transposed vector of each dry monitoring semantic understanding feature vector in the multiple dry monitoring semantic understanding feature vectors to obtain multiple self-attention association matrices; S330, standardizing each self-attention association matrix in the multiple self-attention association matrices to obtain multiple standardized self-attention association matrices; S340, passing each standardized self-attention association matrix in the multiple standardized self-attention association matrices through the Softmax classification function to obtain multiple probability values; S350, weighting each dry monitoring semantic understanding feature vector in the multiple dry monitoring semantic understanding feature vectors with each probability value in the multiple probability values ​​as a weight to obtain the multiple contextual semantic dry monitoring semantic understanding feature vectors; and, S360, cascading the multiple contextual semantic dry monitoring semantic understanding feature vectors to obtain the dry contextual semantic understanding feature vector.

[0067] Specifically, in step S150, after the drying temperature values ​​and vacuum degrees at the plurality of predetermined time points are arranged into a vacuum degree input vector and a drying temperature input vector according to the time dimension, the vacuum degree input vector and the drying temperature input vector are associated and coded to obtain a parameter collaborative input matrix. It should be understood that in the drying process, there is an implicit collaborative correlation between the drying temperature and the vacuum degree. In order to exploit this correlation, in the technical solution of the present application, first, the drying temperature values ​​and vacuum degrees at the plurality of predetermined time points are arranged into a vacuum degree input vector and a drying temperature input vector according to the time dimension to construct the time series discrete distribution of the drying temperature values ​​and the time series discrete distribution of the vacuum degree into the structured vacuum degree input vector and the drying temperature input vector. Then, the vacuum degree input vector and the drying temperature input vector are associated and coded to characterize the correlation between the vacuum degree and the drying temperature values ​​at different time points, thereby obtaining a parameter collaborative input matrix. In a specific example of the present application, the vacuum degree input vector and the drying temperature input vector are associated and coded using the following association formula to obtain a parameter collaborative input matrix; wherein, the formula is: ,in represents the vacuum degree input vector, represents the transposed vector of the vacuum degree input vector, represents the drying temperature input vector, represents the parameter collaborative input matrix, Represents vector multiplication.

[0068] Specifically, in step S160, the parameter collaborative input matrix is ​​passed through the convolutional neural network model as a feature extractor to obtain a parameter collaborative feature matrix. That is, the parameter collaborative input matrix is ​​passed through the convolutional neural network model as a feature extractor to extract feature information related to the collaborative changes of drying parameters, thereby obtaining a parameter collaborative feature matrix. The parameter collaborative feature matrix contains the complex interactive relationship between different parameters in the drying process and has strong expressive power. More specifically, in a specific example of the present application, each layer of the convolutional neural network model used as a feature extractor performs the following on the input data in the forward pass of the layer: convolution processing on the input data to obtain a convolution feature map; pooling along the channel dimension on the convolution feature map to obtain a pooled feature map; and nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the convolutional neural network as a feature extractor is the parameter collaborative input matrix, and the input of the first layer of the convolutional neural network as a feature extractor is the parameter collaborative feature matrix.

[0069] Specifically, in step S170, the responsiveness transfer of the drying context semantic understanding feature vector relative to the parameter collaborative feature matrix is ​​calculated to obtain a classification feature vector. That is, after obtaining the drying context semantic understanding feature vector relative to the parameter collaborative feature matrix, the responsiveness transfer of the drying context semantic understanding feature vector relative to the parameter collaborative feature matrix is ​​further calculated to obtain a classification feature vector. Here, the parameter collaborative feature matrix reflects the complex mutual influence and synergistic effect contained between different drying parameters, while the drying context semantic understanding feature vector can reflect the state of the entire drying process. In the technical solution of the present application, the responsiveness transfer is used to characterize the responsive logical association between the two to optimize the effect of subsequent classification. More specifically, the responsiveness transfer of the drying context semantic understanding feature vector relative to the parameter collaborative feature matrix is ​​calculated using the following formula to obtain a classification feature vector; wherein, the formula is: ,in represents the DRY context semantic understanding feature vector, represents the parameter collaborative feature matrix, represents the classification feature vector.

[0070] Specifically, in step S180, the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the drying temperature value at the current time point is appropriate. That is, in the technical solution of the present application, the classification feature vector is further passed through a classifier to obtain a classification result indicating whether the drying temperature value at the current time point is appropriate. Specifically, the classification feature vector is processed by the classifier using the following formula to obtain a classification result, wherein the formula is: ,in, arrive is the weight matrix, arrive is the bias vector, is a classification feature vector. Specifically, the classifier includes multiple fully connected layers and a Softmax layer cascaded with the last fully connected layer of the multiple fully connected layers. During the classification process of the classifier, the classification feature vector is subjected to multiple fully connected encodings using the multiple fully connected layers of the classifier to obtain an encoded classification feature vector. The encoded classification feature vector is then input into the Softmax layer of the classifier, i.e., the encoded classification feature vector is classified using the Softmax classification function to obtain a classification label. The classifier is a binary classifier that can learn the relationship between the classification feature vector and the actual drying temperature value. In the technical solution of this application, the classification labels include the first label indicating that the drying temperature value at the current time point is appropriate, and the second label indicating that the drying temperature value at the current time point is not appropriate. It is worth noting that the first label p1 and the second label p2 here are not artificially defined concepts. In fact, during the training process, the computer model does not have the concept of "whether the drying temperature value at the current time point is appropriate." Instead, it simply has two classification labels and outputs the probability of the feature under these two classification labels, i.e., the sum of p1 and p2 is one. Therefore, the classification result of whether the drying temperature value at the current time point is appropriate is actually converted into a binary class probability distribution that conforms to natural laws through the classification label. In essence, what is used is the physical meaning of the natural probability distribution of the label, rather than the linguistic text meaning of "whether the drying temperature value at the current time point is appropriate". It should be understood that in the technical solution of the present application, the classification label of the classifier is the control strategy label of whether the drying temperature value at the current time point is appropriate. Therefore, after obtaining the classification result, the drying temperature value at the current time point can be adaptively adjusted based on the classification result. In this way, problems such as product decomposition and loss of crystal water caused by excessively high or low drying temperatures are effectively avoided, thereby improving the purity and quality of electronic-grade lithium hexafluorophosphate products.

[0071] It should be understood that before using the above-mentioned neural network model for inference, it is necessary to train the convolutional neural network model as a filter, the converter-based context encoder, the convolutional neural network model as a feature extractor, and the classifier. In other words, the purification method of electronic-grade lithium hexafluorophosphate of the present application also includes a training module for training the convolutional neural network model as a filter, the converter-based context encoder, the convolutional neural network model as a feature extractor, and the classifier.

[0072] Figure 3 is a flow chart of the training phase in the method for purifying electronic grade lithium hexafluorophosphate according to an embodiment of the present application. As shown in Figure 3, the method for purifying electronic grade lithium hexafluorophosphate according to an embodiment of the present application also includes a training phase, including the steps of: S1110, obtaining training data, the training data including training drying temperature values ​​and training vacuum degrees at multiple predetermined time points within a predetermined time period, training drying monitoring videos of the predetermined time period, and whether the drying temperature value at the current time point is appropriate; S1120, extracting training image frames corresponding to the multiple predetermined time points from the training drying monitoring video as multiple training drying monitoring key frames; S1130, passing the multiple training drying monitoring key frames through the convolutional neural network model as a filter to obtain multiple training drying monitoring semantic understanding feature vectors; S1140, passing the multiple training drying monitoring semantic understanding feature vectors through the context encoder based on the converter to obtain training drying context semantic understanding feature vectors; S1150, arranging the training drying temperature values ​​and training vacuum degrees at the multiple predetermined time points into training vacuum degrees according to the time dimension. After obtaining the vacuum degree input vector and the training drying temperature input vector, the training vacuum degree input vector and the training drying temperature input vector are associatedly encoded to obtain a training parameter collaborative input matrix; S1160, the training parameter collaborative input matrix is ​​passed through the convolutional neural network model as a feature extractor to obtain a training parameter collaborative feature matrix; S1170, the responsiveness transfer of the training drying context semantic understanding feature vector relative to the training parameter collaborative feature matrix is ​​calculated to obtain a training classification feature vector; S1180, the training classification feature vector is periodically reparameterized by the Gumbel normal to obtain a training optimized classification feature vector; S1190, the training optimized classification feature vector is passed through a classifier to obtain a classification loss function value; and, S1200, the convolutional neural network model as a filter, the converter-based context encoder, the convolutional neural network model as a feature extractor and the classifier are trained based on the classification loss function value and through the direction propagation of gradient descent.

[0073] FIG5 is a schematic diagram of the architecture of the training phase in the method for purifying electronic grade lithium hexafluorophosphate according to an embodiment of the present application. As shown in FIG5 , in the method for purifying electronic grade lithium hexafluorophosphate, during the training process, first, training data is obtained, and the training data includes training drying temperature values ​​and training vacuum degrees at multiple predetermined time points within a predetermined time period, training drying monitoring videos of the predetermined time period, and a true value of whether the drying temperature value at the current time point is appropriate; secondly, training image frames corresponding to the multiple predetermined time points are extracted from the training drying monitoring video as multiple training drying monitoring key frames; the multiple training drying monitoring key frames are respectively passed through the convolutional neural network model as a filter to obtain multiple training drying monitoring semantic understanding feature vectors; then, the multiple training drying monitoring semantic understanding feature vectors are passed through the converter-based context encoder to obtain training drying context semantic understanding feature vectors; the training drying temperature values ​​and training vacuum degrees at the multiple predetermined time points are respectively arranged according to the time dimension as training vacuum degree inputs. After the input vector and the training drying temperature input vector are input, the training vacuum input vector and the training drying temperature input vector are associatedly encoded to obtain a training parameter collaborative input matrix; the training parameter collaborative input matrix is ​​passed through the convolutional neural network model as a feature extractor to obtain a training parameter collaborative feature matrix; then, the responsiveness transfer of the training drying context semantic understanding feature vector relative to the training parameter collaborative feature matrix is ​​calculated to obtain a training classification feature vector; the training classification feature vector is periodically reparameterized by the Gumbel normal to obtain a training optimized classification feature vector; the training optimized classification feature vector is passed through a classifier to obtain a classification loss function value; and then, based on the classification loss function value and through the direction propagation of gradient descent, the convolutional neural network model as a filter, the converter-based context encoder, the convolutional neural network model as a feature extractor, and the classifier are trained.

[0074] In the technical solution of the present application, since the classification feature vector is obtained by calculating the responsive transfer of the dry context semantic understanding feature vector relative to the parameter collaborative feature matrix, the feature distribution of the classification feature vector essentially represents the domain transfer feature between the monitoring frame context image semantics expressed by the dry context semantic understanding feature vector and the correlation feature of the drying temperature value and vacuum degree expressed by the parameter collaborative feature matrix. Therefore, due to the difference between the feature domains of the dry context semantic understanding feature vector and the parameter collaborative feature matrix itself, it is easy for a distribution gap to exist in the distribution of the domain transfer feature, which will lead to poor continuity of the overall feature distribution of the classification feature vector and affect the training effect during model training. Based on this, the applicant of the present application has made a correction to the classification feature vector. Perform Gumbel normal periodic reparameterization to obtain the optimized classification feature vector , specifically expressed as: in, represents the training classification feature vector, represents the training optimized classification feature vector, represents the eigenvalues ​​of each position of the training classification feature vector, and are the mean and variance of the eigenvalue set at each position of the training classification feature vector, and , represents the logarithmic function value with base 2, represents the value of the inverse sine function, represents the value of the arccosine function, Here, the Gumbel normal periodic reparameterization is performed by transforming the classification feature vector The eigenvalues ​​at each position Convert it into the angular feature expression of its probability distribution, and introduce random periodic distribution into the normal distribution of the eigenvalue set based on the random periodic operation of Gumbel distribution, so as to obtain a random periodic continuous differentiable approximation of the original feature distribution, thereby improving the optimized classification feature vector through periodic reparameterization of the feature. During training, the gradient of the loss function is dynamically and continuously fluctuated in the back propagation of the model to improve the dynamic applicability of the convolution kernel during the training process, thereby compensating for the impact of the poor continuity of the feature distribution of the classification feature vector on the training effect, such as training speed and the accuracy of the convergence result.

[0075] In summary, the purification method of electronic-grade lithium hexafluorophosphate according to the embodiment of the present application is explained. It uses a deep learning-based neural network model to mine the contextual implicit correlation between the drying temperature and vacuum degree during the purification process, and thereby realizes automatic control of the drying temperature, thereby effectively avoiding product decomposition, loss of crystal water and other problems, and improving the purity and quality of the electronic-grade lithium hexafluorophosphate product.

[0076] Exemplary System: Figure 8 is a block diagram of a system for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application. As shown in Figure 8 , the system 300 for purifying electronic-grade lithium hexafluorophosphate according to an embodiment of the present application includes: an information acquisition module 310; a keyframe extraction module 320; a first convolution module 330; a context encoding module 340; an association encoding module 350; a second convolution module 360; a responsive transfer module 370; and a classification result generation module 380.

[0077] Among them, the information acquisition module 310 is used to obtain the drying temperature value and vacuum degree at multiple predetermined time points within a predetermined time period, as well as the drying monitoring video of the predetermined time period; the key frame extraction module 320 is used to extract the image frames corresponding to the multiple predetermined time points from the drying monitoring video as multiple drying monitoring key frames; the first convolution module 330 is used to pass the multiple drying monitoring key frames through a convolutional neural network model as a filter to obtain multiple drying monitoring semantic understanding feature vectors; the context encoding module 340 is used to pass the multiple drying monitoring semantic understanding feature vectors through a converter-based context encoder to obtain a drying context semantic understanding feature vector; the association encoding module 350 is used to convert the multiple predetermined time points into a drying monitoring key frame. After the drying temperature value and vacuum degree at the intermediate point are arranged as a vacuum degree input vector and a drying temperature input vector according to the time dimension respectively, the vacuum degree input vector and the drying temperature input vector are associated and encoded to obtain a parameter collaborative input matrix; the second convolution module 360 ​​is used to pass the parameter collaborative input matrix through a convolutional neural network model as a feature extractor to obtain a parameter collaborative feature matrix; the responsiveness transfer module 370 is used to calculate the responsiveness transfer of the drying context semantic understanding feature vector relative to the parameter collaborative feature matrix to obtain a classification feature vector; and the classification result generation module 380 is used to pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the drying temperature value at the current time point is appropriate.

[0078] In one example, in the above-mentioned electronic-grade lithium hexafluorophosphate purification system 300, the first convolution module 330 is used to: use the layers of the convolutional neural network model as a filter to respectively perform the following on the input data in the forward pass of the layer: convolution processing on the input data to obtain a convolution feature map; pooling the convolution feature map based on a local feature matrix to obtain a pooled feature map; and, nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the convolutional neural network as a filter is the multiple drying monitoring semantic understanding feature vectors, and the input of the first layer of the convolutional neural network as a filter is the multiple drying monitoring key frames.

[0079] In one example, in the above-mentioned electronic-grade lithium hexafluorophosphate purification system 300, the context encoding module 340 is used to: arrange the multiple drying monitoring semantic understanding feature vectors in one dimension to obtain a global drying monitoring semantic understanding feature vector; calculate the product between the global drying monitoring semantic understanding feature vector and the transposed vector of each drying monitoring semantic understanding feature vector in the multiple drying monitoring semantic understanding feature vectors to obtain multiple self-attention association matrices; standardize each self-attention association matrix in the multiple self-attention association matrices to obtain multiple standardized self-attention association matrices; pass each standardized self-attention association matrix in the multiple standardized self-attention association matrices through a Softmax classification function to obtain multiple probability values; weight each drying monitoring semantic understanding feature vector in the multiple drying monitoring semantic understanding feature vectors using each probability value in the multiple probability values ​​as a weight to obtain the multiple contextual semantic drying monitoring semantic understanding feature vectors; and cascade the multiple contextual semantic drying monitoring semantic understanding feature vectors to obtain the drying contextual semantic understanding feature vector.

[0080] In one example, in the electronic-grade lithium hexafluorophosphate purification system 300, the correlation coding module 350 is configured to perform correlation coding on the vacuum input vector and the drying temperature input vector using the following correlation formula to obtain a parameter coordination input matrix; wherein the formula is: ,in represents the vacuum degree input vector, represents the transposed vector of the vacuum degree input vector, represents the drying temperature input vector, represents the parameter collaborative input matrix, Represents vector multiplication.

[0081] In one example, in the electronic-grade lithium hexafluorophosphate purification system 300, the responsiveness transfer module 370 is configured to calculate the responsiveness transfer of the dry context semantic understanding feature vector relative to the parameter synergy feature matrix using the following formula to obtain a classification feature vector; wherein the formula is: ,in represents the DRY context semantic understanding feature vector, represents the parameter collaborative feature matrix, represents the classification feature vector.

[0082] In summary, the electronic-grade lithium hexafluorophosphate purification system 300 according to the embodiment of the present application is explained. It uses a deep learning-based neural network model to mine the contextual implicit correlation between the drying temperature and the vacuum degree during the purification process, and thereby realizes automatic control of the drying temperature, thereby effectively avoiding problems such as product decomposition and loss of crystal water, thereby improving the purity and quality of the electronic-grade lithium hexafluorophosphate product.

[0083] As described above, the electronic-grade lithium hexafluorophosphate purification system according to the embodiments of the present application can be implemented in various terminal devices. In one example, the electronic-grade lithium hexafluorophosphate purification system according to the embodiments of the present application can be integrated into the terminal device as a software module and / or hardware module. For example, the electronic-grade lithium hexafluorophosphate purification system can be a software module in the terminal device's operating system, or it can be an application developed specifically for the terminal device. Of course, the electronic-grade lithium hexafluorophosphate purification system can also be one of the terminal device's many hardware modules.

[0084] Alternatively, in another example, the electronic-grade lithium hexafluorophosphate purification system and the terminal device may be separate devices, and the electronic-grade lithium hexafluorophosphate purification system may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0085] Exemplary electronic device: Below, an electronic device according to an embodiment of the present application is described with reference to FIG9 .

[0086] FIG9 illustrates a block diagram of an electronic device according to an embodiment of the present application.

[0087] As shown in FIG. 9 , the electronic device 10 includes one or more processors 11 and a memory 12 .

[0088] The processor 11 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

[0089] The memory 12 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), a hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 11 may execute the program instructions to implement the functions of the electronic-grade lithium hexafluorophosphate purification method described in each embodiment of the present application and / or other desired functions. The computer-readable storage medium may also store various contents, such as drying monitoring key frames.

[0090] In one example, the electronic device 10 may further include an input device 13 and an output device 14 , and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0091] The input device 13 may include, for example, a keyboard, a mouse, and the like.

[0092] The output device 14 can output various information to the outside, including classification results, etc. The output device 14 can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and the like.

[0093] Of course, for the sake of simplicity, FIG9 only shows some of the components of the electronic device 10 that are relevant to the present application, omitting components such as a bus, input / output interfaces, etc. In addition, the electronic device 10 may further include any other appropriate components depending on the specific application.

[0094] Exemplary computer program product and computer-readable storage medium: In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to perform the steps in the functions of the purification method of electronic-grade lithium hexafluorophosphate according to various embodiments of the present application described in the above-mentioned "Exemplary Method" section of this specification.

[0095] The computer program product may be written in any combination of one or more programming languages ​​to implement the program code for performing the operations of the embodiments of the present application, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0096] In addition, an embodiment of the present application may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, enable the processor to perform the steps in the functions of the method for purifying electronic-grade lithium hexafluorophosphate according to various embodiments of the present application described in the above-mentioned "Exemplary Method" section of this specification.

[0097] The computer-readable storage medium may be any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may include, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0098] The basic principles of the present application have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in this application are merely illustrative and not restrictive, and it should not be assumed that these advantages, strengths, and effects are required of each embodiment of this application. In addition, the specific details disclosed above are merely illustrative and facilitating understanding, and are not restrictive. The above details do not limit this application to necessarily being implemented using the above specific details.

[0099] The block diagrams of the devices, devices, equipment, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, or configured in any manner. Words such as "include," "comprise," "have," and the like are open-ended words, meaning "including but not limited to," and can be used interchangeably therewith. The words "or" and "and" used herein refer to the words "and / or" and can be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to," and can be used interchangeably therewith.

[0100] It should also be noted that in the apparatus, device, and method of the present application, each component or each step can be decomposed and / or recombined, and such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0101] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0102] The above description has been provided for the purpose of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for purifying electronic grade lithium hexafluorophosphate, characterized in that: include: Acquire drying temperature values ​​and vacuum degrees at a plurality of predetermined time points within a predetermined time period, and a drying monitoring video of the predetermined time period; Extracting the image frames corresponding to the plurality of predetermined time points from the drying monitoring video as a plurality of drying monitoring key frames; Passing the plurality of drying monitoring key frames through a convolutional neural network model as a filter respectively to obtain a plurality of drying monitoring semantic understanding feature vectors; Passing the plurality of dry monitoring semantic understanding feature vectors through a transformer-based context encoder to obtain a dry context semantic understanding feature vector; After arranging the drying temperature values ​​and vacuum degrees at the plurality of predetermined time points into a vacuum degree input vector and a drying temperature input vector according to the time dimension, the vacuum degree input vector and the drying temperature input vector are associatedly encoded to obtain a parameter coordination input matrix; Passing the parameter collaborative input matrix through a convolutional neural network model as a feature extractor to obtain a parameter collaborative feature matrix; Calculating the responsiveness transfer of the dry context semantic understanding feature vector relative to the parameter synergy feature matrix to obtain a classification feature vector; as well as The classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the drying temperature value at the current time point is appropriate.

2. The method for purifying electronic grade lithium hexafluorophosphate according to claim 1, characterized in that: The plurality of drying monitoring key frames are respectively passed through a convolutional neural network model as a filter to obtain a plurality of drying monitoring semantic understanding feature vectors, including: using each layer of the convolutional neural network model as a filter to perform the following on the input data in the forward pass of the layer: Perform convolution on the input data to obtain a convolution feature map; Performing pooling on the convolutional feature map based on a local feature matrix to obtain a pooled feature map; and Performing nonlinear activation on the pooled feature map to obtain an activated feature map; Among them, the output of the last layer of the convolutional neural network as a filter is the multiple drying monitoring semantic understanding feature vectors, and the input of the first layer of the convolutional neural network as a filter is the multiple drying monitoring key frames.

3. The method for purifying electronic grade lithium hexafluorophosphate according to claim 2, characterized in that: The plurality of dry monitoring semantic understanding feature vectors are passed through a transformer-based context encoder to obtain a dry context semantic understanding feature vector, including: Arranging the plurality of drying monitoring semantic understanding feature vectors in one dimension to obtain a global drying monitoring semantic understanding feature vector; Calculating the product of the global dryness monitoring semantic understanding feature vector and the transposed vector of each dryness monitoring semantic understanding feature vector in the multiple dryness monitoring semantic understanding feature vectors to obtain multiple self-attention association matrices; Normalizing each of the plurality of self-attention association matrices to obtain a plurality of standardized self-attention association matrices; Passing each of the plurality of standardized self-attention association matrices through a Softmax classification function to obtain a plurality of probability values; weighting each of the plurality of drying monitoring semantic understanding feature vectors by using each of the plurality of probability values ​​as a weight to obtain the plurality of contextual semantic drying monitoring semantic understanding feature vectors; and The multiple context semantic dry monitoring semantic understanding feature vectors are cascaded to obtain the dry context semantic understanding feature vector.

4. The method for purifying electronic grade lithium hexafluorophosphate according to claim 3, characterized in that: After arranging the drying temperature values ​​and vacuum degrees at the plurality of predetermined time points into a vacuum degree input vector and a drying temperature input vector according to the time dimension, the vacuum degree input vector and the drying temperature input vector are associated and encoded to obtain a parameter coordination input matrix, including: associating and encoding the vacuum degree input vector and the drying temperature input vector according to the following association formula to obtain a parameter coordination input matrix; Wherein, the formula is: in represents the vacuum degree input vector, represents the transposed vector of the vacuum degree input vector, represents the drying temperature input vector, represents the parameter coordination input matrix, Represents vector multiplication.

5. The method for purifying electronic grade lithium hexafluorophosphate according to claim 4, characterized in that: Calculating the responsiveness transfer of the dry context semantic understanding feature vector relative to the parameter collaborative feature matrix to obtain a classification feature vector, comprising: calculating the responsiveness transfer of the dry context semantic understanding feature vector relative to the parameter collaborative feature matrix to obtain a classification feature vector according to the following formula; Wherein, the formula is: in represents the dry context semantic understanding feature vector, represents the parameter synergistic feature matrix, represents the classification feature vector.

6. The method for purifying electronic grade lithium hexafluorophosphate according to claim 5, characterized in that: The method further comprises a training step of: training the convolutional neural network model as a filter, the converter-based context encoder, the convolutional neural network model as a feature extractor, and the classifier; Wherein, the training step includes: Acquire training data, the training data including training drying temperature values ​​and training vacuum degrees at multiple predetermined time points within a predetermined time period, training drying monitoring video of the predetermined time period, and a true value of whether the drying temperature value at a current time point is appropriate; Extracting the training image frames corresponding to the plurality of predetermined time points from the training drying monitoring video as a plurality of training drying monitoring key frames; Passing the plurality of training drying monitoring key frames through the convolutional neural network model as a filter respectively to obtain a plurality of training drying monitoring semantic understanding feature vectors; Passing the plurality of training dry monitoring semantic understanding feature vectors through the transformer-based context encoder to obtain a training dry context semantic understanding feature vector; After arranging the training drying temperature values ​​and the training vacuum degrees at the plurality of predetermined time points into a training vacuum degree input vector and a training drying temperature input vector according to the time dimension, the training vacuum degree input vector and the training drying temperature input vector are associatedly encoded to obtain a training parameter collaborative input matrix; Passing the training parameter collaborative input matrix through the convolutional neural network model as a feature extractor to obtain a training parameter collaborative feature matrix; Calculating the responsiveness transfer of the training dry context semantic understanding feature vector relative to the training parameter collaborative feature matrix to obtain a training classification feature vector; Performing Gumbel normal periodic reparameterization on the training classification feature vector to obtain a training optimized classification feature vector; Passing the training optimized classification feature vector through a classifier to obtain a classification loss function value; and The convolutional neural network model as a filter, the converter-based context encoder, the convolutional neural network model as a feature extractor, and the classifier are trained based on the classification loss function value and propagated in the direction of gradient descent.

7. The method for purifying electronic grade lithium hexafluorophosphate according to claim 6, characterized in that: Performing Gumbel normal periodic reparameterization on the training classification feature vector to obtain a training optimized classification feature vector, comprising: performing Gumbel normal periodic reparameterization on the training classification feature vector according to the following formula to obtain a training optimized classification feature vector; Wherein, the formula is: in, represents the training classification feature vector, represents the training optimized classification feature vector, represents the feature value of each position of the training classification feature vector, and are the mean and variance of the feature value set at each position of the training classification feature vector, and , represents the logarithmic function value with base 2, represents the value of the inverse sine function, represents the value of the arccosine function, Represents the feature value of each position of the optimized training classification feature vector.

8. A purification system for electronic grade lithium hexafluorophosphate, characterized in that: include: An information acquisition module, used to obtain drying temperature values ​​and vacuum degrees at multiple predetermined time points within a predetermined time period, and a drying monitoring video of the predetermined time period; A key frame extraction module, used for extracting image frames corresponding to the plurality of predetermined time points from the drying monitoring video as a plurality of drying monitoring key frames; A first convolution module, used for respectively passing the plurality of drying monitoring key frames through a convolutional neural network model as a filter to obtain a plurality of drying monitoring semantic understanding feature vectors; A context encoding module, configured to pass the plurality of drying monitoring semantic understanding feature vectors through a converter-based context encoder to obtain a drying context semantic understanding feature vector; an associative coding module, for arranging the drying temperature values ​​and vacuum degrees at the plurality of predetermined time points into a vacuum degree input vector and a drying temperature input vector according to the time dimension, and then associatively coding the vacuum degree input vector and the drying temperature input vector to obtain a parameter coordination input matrix; A second convolution module, used for passing the parameter collaborative input matrix through a convolutional neural network model as a feature extractor to obtain a parameter collaborative feature matrix; A responsive transfer module, used for calculating the responsive transfer of the dry context semantic understanding feature vector relative to the parameter synergy feature matrix to obtain a classification feature vector; as well as The classification result generating module is used to pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the drying temperature value at the current time point is appropriate.

9. The electronic grade lithium hexafluorophosphate purification system according to claim 8, characterized in that: The association coding module is used to: perform association coding on the vacuum degree input vector and the drying temperature input vector according to the following association formula to obtain a parameter coordination input matrix; Wherein, the formula is: in represents the vacuum degree input vector, represents the transposed vector of the vacuum degree input vector, represents the drying temperature input vector, represents the parameter coordination input matrix, Represents vector multiplication.

10. The electronic grade lithium hexafluorophosphate purification system according to claim 9, characterized in that: The responsiveness transfer module is used to: calculate the responsiveness transfer of the dry context semantic understanding feature vector relative to the parameter synergy feature matrix using the following formula to obtain a classification feature vector; Wherein, the formula is: in represents the dry context semantic understanding feature vector, represents the parameter synergistic feature matrix, represents the classification feature vector.

Citation Information

Patent Citations

  • Energy management control system for preparing lithium hexafluorophosphate and control method thereof

    CN115079572A

  • Intelligent detection method and system based on neural network model

    CN115099285A

  • Control system and control method of intelligent production line of electronic-grade potassium hydroxide

    CN115202265A

  • Automatic batching system for preparing lithium hexafluorophosphate and batching method thereof

    CN115430344A

  • Production management control system and method for preparing lithium hexafluorophosphate

    CN115586755A

Cited By

  • Traditional Chinese medicine drying quality control method and system based on artificial intelligence

    CN119818970A

  • Traditional Chinese medicine drying quality control method and system based on artificial intelligence

    CN119818970B