Lithium battery moisture prediction method and system based on time and dew point coupling

Abstract

The invention discloses a lithium battery moisture prediction method and system based on time and dew point coupling. The method comprises the following steps: acquiring pole piece processing information and recording the pole piece processing information to a data memory; scanning the identification code of the battery pole piece through identity identification equipment to read the pole piece processing information; inputting the pole piece processing information and the battery pole piece type into a pre-trained lithium battery moisture prediction model to obtain a moisture prediction value of the battery pole piece; when the moisture predicted value exceeds a moisture abnormal threshold value, outputting a battery moisture abnormal early warning and drying the battery pole piece, and adjusting the drying time and the drying temperature of the battery pole piece according to the moisture predicted value of the battery pole piece; according to the method, the battery moisture can be accurately tracked, accurate determination and processing of the battery moisture can be realized, and the situation that the moisture battery cannot be detected is avoided.

Description

Technical Field

[0001] This invention belongs to the field of lithium battery production, specifically relating to a method, system, and storage medium for predicting the moisture content of lithium batteries based on time and dew point coupling. Background Technology

[0002] With the rapid development of the new energy industry, lithium-ion batteries are increasingly widely used in energy storage and power applications due to their high energy density and excellent cycle performance. During battery manufacturing, the moisture content of the electrode sheets is a key parameter affecting the product's electrochemical performance, cycle life, and safety characteristics, requiring strict control. Currently, the industry commonly uses the Karl Fischer moisture testing method for sampling inspection, which involves taking a small number of electrode samples from a specific process for offline destructive testing. While this method meets basic accuracy requirements, it only obtains the moisture data of the tested sample and cannot comprehensively reflect the true moisture distribution of the entire batch of electrodes under different environmental exposure times.

[0003] Because battery electrodes undergo multiple processes during production, the temperature, humidity, and exposure time vary at each stage, resulting in non-uniform moisture absorption across different parts of the electrode. Traditional sampling and testing methods have inherent limitations, such as limited coverage and insufficient representativeness, leading to the risk of missed detections. If electrodes with excessive moisture content are not identified in time and flow into subsequent processes, it may cause side reactions within the battery, resulting in decreased product consistency and even safety hazards. Summary of the Invention

[0004] This invention provides a method and system for predicting the moisture content of lithium batteries based on time and dew point coupling. It can accurately track the moisture content of batteries, accurately determine and process the moisture content of batteries, and avoid situations where moisture-laden batteries cannot be detected.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] The first aspect of this invention provides a method for predicting the moisture content of lithium batteries based on time-dew point coupling, comprising:

[0007] Acquire electrode processing information and record it to a data storage device; the electrode processing information includes the dew point temperature and real-time ambient temperature of the battery processing workshop, as well as the time the battery electrode is placed in the battery processing workshop; the battery electrode is equipped with an identification code;

[0008] The identification code of the battery electrode is scanned by an identification device to read the electrode processing information; the electrode processing information and the battery electrode type are input into a pre-trained lithium battery moisture prediction model to obtain the moisture prediction value of the battery electrode;

[0009] When the predicted moisture value exceeds the moisture abnormality threshold, a battery moisture abnormality warning is output and the battery electrode is dried. The drying time and drying temperature of the battery electrode are adjusted according to the predicted moisture value of the battery electrode.

[0010] This invention deploys environmental sensors and RFID tracking modules to collect and bind real-time information on dew point, temperature, and storage time of individual electrodes, constructing a data foundation for full batch traceability. This completely solves the inherent defects of traditional Karl Fischer sampling, such as incomplete coverage and strong lag. By inputting electrode type and environmental time-series data into a pre-trained model, online intelligent prediction of moisture status is achieved, transforming offline detection into process monitoring.

[0011] Furthermore, the training process of the lithium battery moisture prediction model includes:

[0012] Various types of electrodes were selected and placed in a simulated environment with a specific dew point temperature. At fixed time intervals, the moisture content of the electrodes was measured using the Karl Fischer moisture test method to obtain moisture monitoring values, and the temperature monitoring information of the simulated environment was recorded.

[0013] Given the selected electrode type and dew point temperature, a simulation fitting function is obtained by fitting the relationship curve between the moisture monitoring value and the resting time; the resting time is randomly sampled to obtain the resting sampling time, and the resting sampling time is input into the simulation fitting function to obtain the moisture sampling value.

[0014] The dew point temperature, the time of rest sampling, and the temperature monitoring information are strictly aligned according to the timestamp. Z-score standardization is used to eliminate the dimensional differences of electrode type, dew point temperature, time of rest sampling, and temperature monitoring information. The samples are then spliced ​​along the feature channel dimension to form training samples. The moisture sampling value is used as the true label of the training samples.

[0015] The training samples are input into the 1D-CNN model to obtain moisture prediction training values; the training loss is calculated based on the moisture prediction training values ​​and the true labels; the parameters of the 1D-CNN model are optimized based on the training loss, and the training process of the 1D-CNN model is repeated iteratively until the training loss converges and the final lithium battery moisture prediction model is output.

[0016] This invention conducts long-term electrode moisture absorption experiments under various dew point conditions in a simulated environment, constructing a high-confidence mapping dataset of environment-time-moisture using the Karl Fischer method as the truth source. Z-score normalization is employed to eliminate dimensional differences in electrode type, dew point temperature, storage time, and temperature monitoring information. High-dimensional time-series training samples are then formed by concatenating feature channels, effectively solving the feature coupling problem of heterogeneous data sources. A hybrid modeling approach embeds prior physical knowledge of the battery moisture absorption mechanism into a data-driven framework, allowing the 1D-CNN model to learn nonlinear relationships while being constrained by physical laws, resulting in predictions that combine statistical accuracy and interpretability.

[0017] Furthermore, a simulation fitting function is obtained by fitting a curve to the relationship between moisture monitoring values ​​and storage time, specifically including:

[0018]

[0019] In the formula, This is a moisture monitoring value; This is the time for suspension; and These are the equation coefficients of the simulation fitting function; and The power exponent of the equation for the simulation fitting function; This represents the initial moisture content of the electrode.

[0020] In this invention, the simulation fitting function uses two power terms of different orders to characterize the dual mechanisms of rapid surface penetration in the initial stage of water adsorption and slow internal diffusion in the later stage, while the constant term W0 distinguishes the differences in the initial water baseline. The simulation fitting function has a stronger fitting ability to the actually measured S-shaped growth curve, providing high-quality pseudo-label data for subsequent deep learning and reducing the learning difficulty and overfitting risk of the neural network.

[0021] Furthermore, the training loss is calculated based on the moisture prediction training values ​​and the true labels; specifically including:

[0022]

[0023]

[0024]

[0025] In the formula, For training loss, , and To set the weight parameters; Main loss item; The mean squared error between the training values ​​and the true labels for moisture prediction; This results in a loss of time-series consistency. This is a real label; Training values ​​for moisture prediction; is the error threshold; N is the number of input training samples.

[0026] The main loss term of this invention is when the error is less than a threshold. It exhibits smooth characteristics to stabilize the gradient, greater than Temporal consistency loss enhances resistance to outliers. By constraining the difference between the derivatives of predicted values ​​and true labels, temporal consistency loss forces the model to learn the monotonically continuous law of water growth, avoiding predictions that violate physical laws. Weighting coefficients. , and Achieving a dynamic balance among various loss terms ensures both numerical accuracy and enhances the rationality of dynamic characteristics, enabling lithium battery moisture prediction to achieve the optimal trade-off between convergence speed, anti-interference capability, and physical interpretability.

[0027] Furthermore, the electrode processing information and battery electrode type are input into a pre-trained lithium battery moisture prediction model to obtain the predicted moisture value of the battery electrode, specifically including:

[0028] The lithium battery moisture prediction model includes an input layer, a first feature extraction unit, a second feature extraction unit, a Flatten layer, a fully connected layer, and an output layer connected in sequence.

[0029] Electrode processing information and battery electrode type are concatenated to obtain electrode state features. These electrode state features are input to the lithium battery moisture prediction model from the input layer, and then processed sequentially through the first feature extraction unit, the second feature extraction unit, the Flatten layer, and the fully connected layer. Finally, the predicted moisture value of the battery electrode is output through the output layer.

[0030] This invention combines electrode processing information with battery electrode type to form electrode state features, integrating multi-dimensional electrode-related information to avoid the feature limitations of single-dimensional information. This allows the model input to better reflect the actual influencing factors of electrode moisture. Through a layered and progressive feature processing flow, the model progressively mines the correlation between features and moisture, thereby accurately capturing the nonlinear dynamic evolution of moisture growth relative to environmental parameters; this significantly improves the physical accuracy and environmental adaptability of moisture prediction.

[0031] Furthermore, the local temporal primary features are obtained by performing linear transformation and nonlinear activation on the polarimetric state features through the convolutional layer in the first feature extraction unit, and intermediate features are obtained by downsampling the local temporal primary features using the pooling layer in the first feature extraction unit.

[0032] In this invention, the first feature extraction unit uses convolutional layers to achieve linear transformation and nonlinear activation, which can accurately capture local temporal correlations in the polarimetric state features. The activation operation also enhances the model's nonlinear fitting ability. The downsampling of the pooling layer removes redundant information in the primary local temporal features, simplifies the feature dimensions, and improves feature processing efficiency.

[0033] Furthermore, the convolutional layer in the second feature extraction unit captures long-term dependencies of intermediate features to obtain coupled features, and the pooling layer in the second feature extraction unit downsamples the coupled features to obtain key temporal features.

[0034] In this invention, the convolutional layer of the second feature extraction unit captures long-term dependencies in intermediate features, breaking the limitations of local features and integrating them to form coupled features that fuse long-term and short-term correlations, making the features more closely match the temporal variation patterns of electrode moisture. Subsequent pooling layers further downsample, refining key temporal features, weakening interference from irrelevant information, and strengthening the representational ability of core features.

[0035] Furthermore, key time-series features are input into the Flatten layer to obtain a global feature vector. The global feature vector is then compressed through a fully connected layer to obtain a prediction factor vector. The prediction factor vector is then input into the output layer to perform linear regression and output the predicted moisture value of the battery electrode.

[0036] In this invention, the Flatten layer transforms key temporal features into global feature vectors, achieving dimensional integration and flattening of temporal features, which facilitates feature processing by the fully connected layer. Feature compression in the fully connected layer can extract predictive factor vectors highly correlated with water content, while the linear regression of the output layer adapts to the prediction requirements of continuous water content values, accurately mapping the correlation between core features and water content values, and improving prediction accuracy.

[0037] Furthermore, the drying time and temperature of the battery electrode are adjusted according to the predicted moisture content of the battery electrode, specifically including:

[0038]

[0039]

[0040] In the formula, This is the drying time adjustment value; This is the predicted moisture value. The target moisture content; The efficiency coefficient of the electrode material; This is the drying temperature adjustment value; The baseline drying temperature; Basic time margin; Standard drying temperature; To set the weight parameters.

[0041] This invention dynamically adjusts the time and temperature parameters of the drying process based on predicted values. The adaptive control mechanism can implement differentiated treatment strategies for different electrode types and different degrees of exceeding the standard. This avoids energy waste and material damage caused by over-drying, while ensuring that the drying effect reaches the target moisture content, thus achieving refined and intelligent setting of process parameters.

[0042] A second aspect of the present invention provides a lithium battery moisture prediction system based on time-dew point coupling, comprising:

[0043] The data monitoring module acquires electrode processing information and records it to the data storage device; the electrode processing information includes the dew point temperature and real-time ambient temperature of the battery processing workshop, as well as the time the battery electrode is placed in the battery processing workshop.

[0044] The moisture prediction module scans the identification code of the battery electrode using an identification device to read the electrode processing information; it then inputs the electrode processing information and the battery electrode type into a pre-trained lithium battery moisture prediction model to obtain the predicted moisture value of the battery electrode.

[0045] The early warning module outputs a battery moisture abnormality warning when the predicted moisture value exceeds the moisture abnormality threshold.

[0046] The control module dries the battery electrode when the predicted moisture value exceeds the abnormal moisture threshold; and adjusts the drying time and temperature of the battery electrode according to the predicted moisture value of the battery electrode.

[0047] This invention utilizes a modular design to break down complex technical processes into four independent yet collaborative functional units: data monitoring, moisture prediction, early warning, and control. This facilitates phased implementation and maintenance upgrades in actual production lines. The hardware combination of the RFID tracking module and data storage device establishes a digital identity file for each electrode, supporting quality traceability throughout its entire lifecycle. The separate design of the early warning and control modules ensures both the timeliness of risk notification and the accuracy of process intervention.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] This invention combines an RFID tracking module with a multi-parameter real-time acquisition system to upgrade the detection target from discrete samples to the entire batch of electrode sheets. It enables the tracking of the moisture status of each battery electrode sheet throughout the entire process from coating and rolling to assembly, avoiding internal side reactions, decreased product consistency, and safety hazards that may be caused by electrode sheets with excessive moisture flowing into subsequent processes. This fundamentally improves the tightness of the quality control system.

[0050] When the predicted moisture content exceeds a preset threshold, the system automatically triggers an early warning mechanism to promptly intercept abnormal batches and prevent substandard products from spreading downstream. More importantly, the system dynamically adjusts the time and temperature parameters of the drying process based on the predicted value, transforming moisture control from "post-event detection" to "pre-event prevention" and "in-process optimization." This adaptive control mechanism can implement differentiated treatment strategies for different electrode types and varying degrees of moisture exceeding the threshold, avoiding energy waste and material damage caused by over-drying while ensuring that the drying effect reaches the target moisture content, achieving refined and intelligent setting of process parameters. Attached Figure Description

[0051] Figure 1 This is a flowchart of the lithium battery moisture prediction method provided in Embodiment 1 of the present invention;

[0052] Figure 2 This is a graph showing the relationship between the moisture content of the electrode and the storage time when the dew point temperature is 0°C, as provided in Embodiment 1 of the present invention.

[0053] Figure 3 This is a graph showing the relationship between the moisture content of the electrode and the storage time when the dew point temperature is -10℃, as provided in Embodiment 1 of the present invention.

[0054] Figure 4 This is a graph showing the relationship between the moisture content of the electrode and the storage time when the dew point temperature is -20℃, as provided in Embodiment 1 of the present invention.

[0055] Figure 5 This is a graph showing the relationship between the moisture content of the electrode and the storage time when the dew point temperature is -30℃, as provided in Embodiment 1 of the present invention.

[0056] Figure 6 This is a structural diagram of the lithium battery moisture prediction system provided in Embodiment 1 of the present invention. Detailed Implementation

[0057] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0058] Example 1

[0059] like Figure 1 As shown, this embodiment provides a method for predicting the moisture content of lithium batteries based on time-dew point coupling, including:

[0060] Acquire electrode processing information and record it to a data storage device; the electrode processing information includes the dew point temperature and real-time ambient temperature of the battery processing workshop, as well as the time the battery electrode is placed in the battery processing workshop;

[0061] The battery electrode is equipped with an identification code; the identification code of the battery electrode is scanned and the electrode processing information is read by an identification device; in this embodiment, the identification device is an RFID reader / writer; the identification code of the battery electrode is an RFID electronic tag; in addition, the identification device can also be set as a barcode identification device, a QR code identification device, and a data matrix code reading device. Each type of device can specifically complete the reading of the electrode identification code and accurately obtain the corresponding processing information.

[0062] This embodiment deploys environmental sensors and an RFID tracking module to collect and bind real-time information on the dew point, temperature, and storage time of individual electrodes, constructing a data foundation for full batch traceability. This completely solves the inherent defects of traditional Karl Fischer sampling, such as incomplete coverage and strong lag. By inputting electrode type and environmental time-series data into a pre-trained model, online intelligent prediction of moisture status is achieved, transforming offline detection into process monitoring.

[0063] The electrode processing information and battery electrode type are input into a pre-trained lithium battery moisture prediction model to obtain the predicted moisture value of the battery electrode, specifically including:

[0064] The lithium battery moisture prediction model includes an input layer, a first feature extraction unit, a second feature extraction unit, a Flatten layer, a fully connected layer, and an output layer.

[0065] The input layer receives the electrode state features, and the convolutional layer in the first feature extraction unit performs linear transformation and nonlinear activation on the electrode state features to obtain local temporal primary features. The pooling layer in the first feature extraction unit then downsamples the local temporal primary features to obtain intermediate features.

[0066] The convolutional layer in the second feature extraction unit captures long-term dependencies in intermediate features to obtain coupled features, and the pooling layer in the second feature extraction unit downsamples the coupled features to obtain key temporal features.

[0067] Key time-series features are input into the Flatten layer to obtain a global feature vector. The global feature vector is then compressed through a fully connected layer to obtain a prediction factor vector. The prediction factor vector is then input into the output layer to perform linear regression and output the predicted moisture value of the battery electrode.

[0068] In this embodiment, electrode processing information and battery electrode type are concatenated to form electrode state features, which are then input into a multi-layer model. This integrates multi-dimensional related information, avoids the feature defects of single information, and the layered and progressive processing flow can also explore the intrinsic relationship between features and moisture content layer by layer. The first feature extraction unit captures local temporal primary features through convolutional layers and enhances the model's nonlinear fitting ability. The pooling layer downsampling removes redundancy and simplifies dimensions, improving processing efficiency. The convolutional layer of the second feature extraction unit can break through local limitations to capture long-term dependencies, forming coupled features that integrate long-term and short-term correlations. The pooling layer further refines key temporal features and weakens the interference of irrelevant information. The Flatten layer integrates and flattens the dimensions of temporal features, facilitating subsequent processing. The fully connected layer compresses and refines highly correlated predictive factor vectors, and the linear regression of the output layer accurately adapts to the prediction requirements of continuous moisture values. The layered connection effectively improves the accuracy of battery electrode moisture prediction.

[0069] When the predicted moisture content exceeds the moisture anomaly threshold, a battery moisture anomaly warning is issued, and the battery electrodes are dried. The drying time and temperature of the battery electrodes are adjusted according to the predicted moisture content, specifically including:

[0070]

[0071]

[0072] In the formula, This is the drying time adjustment value; This is the predicted moisture value. The target moisture content; The efficiency coefficient of the electrode material; This is the drying temperature adjustment value; The baseline drying temperature; Basic time margin; Standard drying temperature; To set the weight parameters.

[0073] Based on the predicted values, the time and temperature parameters of the drying process are dynamically adjusted. The adaptive control mechanism can implement differentiated treatment strategies for different electrode types and different degrees of exceeding the standard. This avoids energy waste and material damage caused by over-drying, while ensuring that the drying effect reaches the target moisture content, thus realizing the refinement and intelligence of process parameter setting.

[0074] The training process of the lithium battery moisture prediction model includes:

[0075] Various types of electrodes were selected and placed in a simulated environment with a specific dew point temperature. At fixed time intervals, the moisture content of the electrodes was measured using the Karl Fischer moisture test method to obtain moisture monitoring values, and the temperature monitoring information of the simulated environment was recorded.

[0076] In this embodiment, the positive electrode sheet after rolling of the lithium iron phosphate system is selected as the sample for data collection. The sample is the coated electrode sheet.

[0077] The coated electrode sheets were subjected to initial moisture content testing, with four experimental groups: Group 1, Group 2, Group 3, and Group 4. A Karl Fischer moisture analyzer was used, with a testing temperature of 150℃ to 180℃ and a testing time of 180s to 200s. After the test, the initial moisture content of the electrode sheets was recorded: W1 for Group 1, W2 for Group 2, W3 for Group 3, and W4 for Group 4.

[0078] like Figures 2 to 5 As shown, in this embodiment, the samples were placed in a simulation environment with dew point temperatures set as follows: Group 1: -30℃, Group 2: -20℃, Group 3: -10℃, and Group 4: 0℃. Every hour, a portion of the electrode was removed for water content testing. The moisture content of the sample electrodes was measured using a Karl Fischer moisture analyzer. The moisture content at each time point was recorded.

[0079] Given the selected electrode type and dew point temperature, a simulation fitting function is obtained by fitting a curve to the relationship between moisture monitoring values ​​and storage time; the formula is as follows:

[0080]

[0081] In the formula, This is a moisture monitoring value; This is the time for suspension; and These are the equation coefficients of the simulation fitting function; and The power exponent of the equation for the simulation fitting function; This represents the initial moisture content of the electrode.

[0082] The simulation fitting function uses two power terms of different orders to characterize the dual mechanism of rapid surface penetration in the initial stage of water adsorption and slow internal diffusion in the later stage, while the constant term W0 distinguishes the differences in the initial water baseline. The simulation fitting function has a stronger fitting ability to the actually measured S-shaped growth curve, providing high-quality pseudo-label data for subsequent deep learning and reducing the learning difficulty and overfitting risk of the neural network.

[0083] The moisture values ​​used for fitting under the dew point environment in Group 1 With the shelving time The equation is:

[0084]

[0085] The moisture values ​​used for fitting under the dew point environment in Group 2 With the shelving time The equation is:

[0086]

[0087] The moisture values ​​used for fitting under the dew point environment in Group 3 With the shelving time The equation is:

[0088]

[0089] The moisture values ​​used for fitting under the dew point environment in Group 4 With the shelving time The equation is:

[0090]

[0091] In the formula, to The initial moisture content of the electrodes in groups 1 to 4 are respectively.

[0092] Randomly sample the settling time to obtain the settling sampling time, and input the settling sampling time into the simulation fitting function to obtain the moisture sampling value;

[0093] The dew point temperature, sampling time, and temperature monitoring information are strictly aligned according to the timestamp. Z-score standardization is used to eliminate the dimensional differences between electrode type, dew point temperature, sampling time, and temperature monitoring information. The samples are then stitched together along the feature channel dimension to form training samples. The moisture sampling value is used as the true label of the training samples.

[0094] Long-term electrode moisture absorption experiments were conducted under various dew point conditions in a simulated environment. Using the Karl Fischer method as the ground truth, a high-confidence mapping dataset of environment-time-moisture was constructed. Z-score normalization was used to eliminate dimensional differences in electrode type, dew point temperature, storage time, and temperature monitoring information. Then, high-dimensional time-series training samples were formed by concatenating feature channels, effectively solving the feature coupling problem of heterogeneous data sources.

[0095] The training samples are input into the 1D-CNN model to obtain moisture prediction training values; the training loss is calculated based on the moisture prediction training values ​​and the true labels; the formula is as follows:

[0096]

[0097]

[0098]

[0099] In the formula, For training loss, , and To set the weight parameters; Main loss item; The mean squared error between the training values ​​and the true labels for moisture prediction; This results in a loss of time-series consistency. This is a real label; Training values ​​for moisture prediction; is the error threshold; N is the number of input training samples.

[0100] The parameters of the 1D-CNN model are optimized based on the training loss. The training process of the 1D-CNN model is repeated iteratively until the training loss converges and the final lithium battery moisture prediction model is output.

[0101] The training process of the lithium battery moisture prediction model simulates the multi-scale observation logic of manual analysis of moisture evolution: it pays attention to both the impact of recent environmental changes and the cumulative consequences of historical exposure; thus accurately capturing the nonlinear dynamic evolution law of moisture growth relative to environmental parameters; and greatly improving the physical accuracy and environmental adaptability of moisture prediction.

[0102] Example 2

[0103] like Figure 6 As shown, this embodiment provides a lithium battery moisture prediction system based on time and dew point coupling. The lithium battery moisture prediction system is used to execute the lithium battery moisture prediction method described in Embodiment 1. The lithium battery moisture prediction system includes:

[0104] The data monitoring module acquires electrode processing information and records it to the data storage device; the electrode processing information includes the dew point temperature and real-time ambient temperature of the battery processing workshop, as well as the time the battery electrode is placed in the battery processing workshop; the battery electrode is equipped with an identification code.

[0105] The moisture prediction module scans the identification code of the battery electrode using an identification device to read the electrode processing information; it then inputs the electrode processing information and the battery electrode type into a pre-trained lithium battery moisture prediction model to obtain the predicted moisture value of the battery electrode.

[0106] The early warning module outputs a battery moisture abnormality warning when the predicted moisture value exceeds the moisture abnormality threshold.

[0107] The control module dries the battery electrode when the predicted moisture value exceeds the abnormal moisture threshold; and adjusts the drying time and temperature of the battery electrode according to the predicted moisture value of the battery electrode.

[0108] The modular design breaks down the complex technical process into four independent yet collaborative functional units: data monitoring, moisture prediction, early warning, and control. This facilitates phased implementation and maintenance upgrades on the actual production line. The hardware combination of the RFID tracking module and data storage device establishes a digital identity file for each electrode, supporting quality traceability throughout its entire lifecycle. The separate design of the early warning and control modules ensures both the timeliness of risk notification and the accuracy of process intervention.

[0109] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0110] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0111] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0112] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0113] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting the moisture content of lithium batteries based on time-dew point coupling, characterized in that, include: Acquire electrode processing information and record it to a data storage device; the electrode processing information includes the dew point temperature and real-time ambient temperature of the battery processing workshop, as well as the time the battery electrode is placed in the battery processing workshop; The battery electrode processing information is read by scanning the identification code of the electrode using an identification device; The electrode processing information and battery electrode type are input into a pre-trained lithium battery moisture prediction model to obtain the predicted moisture value of the battery electrode. When the predicted moisture value exceeds the moisture abnormality threshold, a battery moisture abnormality warning is output and the battery electrode is dried. The drying time and drying temperature of the battery electrode are adjusted according to the predicted moisture value of the battery electrode.

2. The lithium battery moisture prediction method according to claim 1, characterized in that, The training process of the lithium battery moisture prediction model includes: Various types of electrodes were selected and placed in a simulated environment with a specific dew point temperature. Moisture content of the electrodes was tested at fixed time intervals to obtain moisture monitoring values, and temperature monitoring information of the simulated environment was recorded. Given the selected electrode type and dew point temperature, a simulation fitting function is obtained by fitting the relationship curve between the moisture monitoring value and the resting time; the resting time is randomly sampled to obtain the resting sampling time, and the resting sampling time is input into the simulation fitting function to obtain the moisture sampling value. Z-score standardization was used to eliminate the dimensional differences in electrode type, dew point temperature, resting sampling time, and temperature monitoring information, and the samples were spliced ​​along the feature channel dimension to form training samples; the moisture sampling value was used as the true label of the training samples. The training samples are input into the 1D-CNN model to obtain moisture prediction training values; the training loss is calculated based on the moisture prediction training values ​​and the true labels; the parameters of the 1D-CNN model are optimized based on the training loss, and the training process of the 1D-CNN model is repeated iteratively until the training loss converges and the final lithium battery moisture prediction model is output.

3. The lithium battery moisture prediction method according to claim 2, characterized in that, The simulation fitting function includes: ； In the formula, This is a moisture monitoring value; This is the time for suspension; and These are the equation coefficients of the simulation fitting function; and The power exponent of the equation for the simulation fitting function; This represents the initial moisture content of the electrode.

4. The lithium battery moisture prediction method according to claim 2, characterized in that, The training loss is calculated based on the moisture prediction training values ​​and the true labels; specifically... include: ； ； ； In the formula, For training loss, , and To set the weight parameters; Main loss item; The mean squared error between the training values ​​and the true labels for moisture prediction; This results in a loss of time-series consistency. This is a real label; Training values ​​for moisture prediction; is the error threshold; N is the number of input training samples.

5. The method for predicting the moisture content of lithium batteries according to claim 1, characterized in that, The electrode processing information and battery electrode type are input into a pre-trained lithium battery moisture prediction model to obtain the predicted moisture value of the battery electrode, specifically including: The lithium battery moisture prediction model includes an input layer, a first feature extraction unit, a second feature extraction unit, a Flatten layer, a fully connected layer, and an output layer connected in sequence. Electrode processing information and battery electrode type are concatenated to obtain electrode state features. These electrode state features are input to the lithium battery moisture prediction model from the input layer, and then processed sequentially through the first feature extraction unit, the second feature extraction unit, the Flatten layer, and the fully connected layer. Finally, the predicted moisture value of the battery electrode is output through the output layer.

6. The lithium battery moisture prediction method according to claim 5, characterized in that, The first feature extraction unit uses a convolutional layer to perform linear transformation and nonlinear activation on the polarization state features to obtain local temporal primary features, and then uses a pooling layer in the first feature extraction unit to downsample the local temporal primary features to obtain intermediate features.

7. The lithium battery moisture prediction method according to claim 6, characterized in that, The convolutional layer in the second feature extraction unit captures long-term dependencies in intermediate features to obtain coupled features, and the pooling layer in the second feature extraction unit downsamples the coupled features to obtain key temporal features.

8. The method for predicting the moisture content of lithium batteries according to claim 7, characterized in that, Key time-series features are input into the Flatten layer to obtain a global feature vector. The global feature vector is then compressed through a fully connected layer to obtain a prediction factor vector. The prediction factor vector is then input into the output layer to perform linear regression and output the predicted moisture value of the battery electrode.

9. The method for predicting the moisture content of lithium batteries according to claim 1, characterized in that, Adjusting the drying time and temperature of the battery electrodes based on the predicted moisture content of the electrodes specifically includes: ； ； In the formula, This is the drying time adjustment value; This is the predicted moisture value. The target moisture content; The efficiency coefficient of the electrode material; This is the drying temperature adjustment value; The baseline drying temperature; Basic time margin; Standard drying temperature; To set the weight parameters.

10. A lithium battery moisture prediction system based on time-dew point coupling, characterized in that, include: The data monitoring module acquires electrode processing information and records it to the data storage device; the electrode processing information includes the dew point temperature and real-time ambient temperature of the battery processing workshop, as well as the time the battery electrode is placed in the battery processing workshop. The moisture prediction module scans the identification code of the battery electrode to read the electrode processing information using an identification device; The electrode processing information and battery electrode type are input into a pre-trained lithium battery moisture prediction model to obtain the predicted moisture value of the battery electrode. The early warning module outputs a battery moisture abnormality warning when the predicted moisture value exceeds the moisture abnormality threshold. The control module dries the battery electrode when the predicted moisture value exceeds the abnormal moisture threshold; and adjusts the drying time and temperature of the battery electrode according to the predicted moisture value of the battery electrode.

Images