A smart execution system for adaptive doping processes

By using an intelligent execution system for adaptive doping processes and spectral data processing technology, dynamic adjustment of the lithium iron phosphate doping process was achieved, solving the problems of material consistency and performance stability, and improving the accuracy and reliability of doping control.

CN123120159APending Publication Date: 2026-11-17SHENZHEN WARRANT NEW ENERGY CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611628085.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-21
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies in lithium iron phosphate production cannot effectively utilize spectral data to dynamically adjust the doping process, making it difficult to guarantee material consistency and performance stability.

Method used

The intelligent execution system employing adaptive doping technology uses spectral data processing to divide the main peak core region and inter-peak transition region, calculate the feature matrix and correlation matrix, obtain the raw material state coordinates, and combine them with parameters such as the purity of metal ion compounds to control the doping amount and achieve precise doping.

Benefits of technology

This improved the control precision and reliability of the lithium iron phosphate doping process, and enhanced the material consistency and electrochemical performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123120159A_ABST
    Figure CN123120159A_ABST
Patent Text Reader

Abstract

This invention provides an intelligent execution system for an adaptive doping process. It collects spectral data from K sampling locations of lithium iron phosphate raw materials to form a sample set; identifies the main peak position in the standard phase spectrum, divides the main peak core region and inter-peak transition region, and generates first and second feature matrices; obtains the internal correlation matrix and cross-domain correlation matrix from row covariance and cross-covariance, respectively; constructs an orthogonal basis by eigenvalue decomposition of the internal correlation matrix; projects and aggregates the cross-domain correlation matrix by row direction; and obtains L-dimensional raw material state coordinates by combining the row mean offset vector; uses the pseudo-inverse of the sensitivity matrix to map the difference between the target and the current state to obtain the theoretical doping adjustment amount; and calculates the actual doping amount according to the total raw material mass, effective compound purity, metal element mass fraction, and non-negative constraints to complete the feeding control; and sequentially performs carbon-adding grinding, spray drying, high-temperature sintering solid-state reaction, and pulverization, demagnetization, and grading to obtain the finished lithium iron phosphate product.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application belongs to the field of materials processing, and in particular relates to an intelligent execution system for adaptive doping processes. Background Technology

[0002] Lithium iron phosphate (LiFePO4) is a mainstream cathode material for lithium-ion batteries, boasting advantages such as high safety, long cycle life, and low cost. However, its inherently low electronic conductivity and lithium-ion diffusion rate limit its application in high-rate charge-discharge scenarios. To address this issue, the industry typically employs elemental doping to modify lithium iron phosphate. The doping elements usually include at least one of Mg (magnesium), Ti (titanium), V (vanadium), Mn (manganese), and Nb (niobium). The introduction of these metal ions can induce moderate lattice distortion, broaden lithium-ion transport channels, or alter the band structure of the material, thereby improving the electrochemical performance of lithium iron phosphate. Examples include a method for preparing nano-lithium iron phosphate cathode materials for lithium-ion batteries disclosed in patent CN103474661A, and a lithium iron phosphate composite material and its preparation method disclosed in CN118899425B. In lithium iron phosphate production, to ensure the doping process closely reflects the actual properties of the raw materials, spectroscopic detection technology can be used to perform multi-location sampling analysis of the lithium iron phosphate raw materials to be doped. Spectroscopic techniques (such as infrared spectroscopy, Raman spectroscopy, or ICP spectroscopy) are used to detect the phase composition of the synthesized or doped lithium iron phosphate final product to confirm the state of the dopant elements. However, due to fluctuations in upstream processes, different batches or different locations within the same batch of raw materials to be doped often exhibit slight differences in physicochemical state. Existing technologies typically use static doping formulations with fixed proportions for elements such as Mg, Ti, V, Mn, or Nb, without converting real-time acquired spectral data into dynamic adjustment parameters for upstream processes. Spectral curves contain multi-dimensional information about the material's lattice vibrations and chemical bonding, serving as crucial data indicating the current phase state of the raw material. By analyzing the morphology, characteristic peak positions, and distribution of this spectral data, technicians can assess the current state of the raw material. As the battery manufacturing industry's requirements for material consistency and performance stability continue to evolve, raw material state monitoring and doping control methods based on spectral analysis are gradually becoming standard practices in production. How to utilize spectral data to achieve scientific control of the doping process is a topic of discussion in the field of materials processing. Summary of the Invention

[0003] According to one aspect of the present invention, an intelligent execution system for adaptive doping processes is proposed, which utilizes spectral data to achieve scientific control of the doping process, comprising the following modules: The calculation module is used to acquire spectral data from K sampling locations of lithium iron phosphate raw materials to form a sample set; identify the main peak position according to the phase standard spectrum, and divide the single spectrum into the main peak core region and the inter-peak transition region using the unified wavenumber sampling point index as the horizontal axis reference; calculate the second-order difference value in each region, and take the local extreme point as the initial feature point; determine the weight according to the signal amplitude and the absolute value of the second-order difference, select the spectral amplitude feature value in the preset N main peak core alignment windows and M inter-peak transition alignment windows, and form the first and second feature vectors according to the fixed spectral position order; The module is used to traverse the sample set to generate aligned first and second feature matrices; calculate the row covariance of the first feature matrix to obtain the internal correlation matrix; calculate the cross-domain correlation matrix by calculating the cross-covariance of the two matrices; perform eigenvalue decomposition on the internal correlation matrix and extract the eigenvectors corresponding to the top L largest eigenvalues ​​to form an orthogonal basis; project the cross-domain correlation matrix onto the orthogonal basis to obtain the projected correlation matrix, and aggregate it according to the row weight vector of the preset inter-peak feature direction; combine it with the row mean offset vector of the two feature matrices to output L-dimensional raw material state coordinates. The execution module is used to multiply the pseudo-inverse matrix of the preset sensitivity matrix by the difference vector between the target state coordinates and the raw material state coordinates to obtain the theoretical doping adjustment amount; and combine the total mass of the raw material, the effective purity of the metal ion compound, the mass fraction of the metal element in the compound and the non-negative constraint to calculate the actual doping amount and control the feeding mechanism to release the metal dopant. The finished product output module is used to sequentially process the raw materials after adding the metal dopant into carbon mixing and grinding, spray drying and molding, high-temperature sintering solid-phase reaction in an inert atmosphere, and to break up, crush and demagnetize the sintered materials to obtain lithium iron phosphate finished products.

[0004] According to another aspect of the present invention, a smart execution method for adaptive doping processes is proposed, comprising the following steps: A sample set is formed by acquiring spectral data from K sampling locations of lithium iron phosphate raw materials; the main peak position is identified according to the phase standard spectrum, and the single spectrum is divided into the main peak core region and the inter-peak transition region using the unified wavenumber sampling point index as the abscissa reference; the second-order difference value is calculated in each region, and the local extreme point is taken as the initial feature point; the weight is determined according to the signal amplitude and the absolute value of the second-order difference, and the spectral amplitude feature values ​​are selected in the preset N main peak core alignment windows and M inter-peak transition alignment windows, and the first and second feature vectors are formed according to the fixed spectral position order; The sample set is traversed to generate aligned first and second feature matrices; the row covariance of the first feature matrix is ​​calculated to obtain the internal correlation matrix; the cross-domain correlation matrix is ​​obtained by calculating the cross-covariance of the two matrices; the internal correlation matrix is ​​eigenvalued and the eigenvectors corresponding to the top L largest eigenvalues ​​are extracted to form an orthogonal basis; the cross-domain correlation matrix is ​​projected onto the orthogonal basis to obtain the projected correlation matrix, and the matrix is ​​aggregated according to the row direction of the preset inter-peak feature direction weight vector. Combined with the row mean offset vector of the two feature matrices, the L-dimensional raw material state coordinates are output. The theoretical doping adjustment amount is obtained by left-multiplying the pseudo-inverse matrix of the preset sensitivity matrix by the difference vector between the target state coordinate and the raw material state coordinate; and by combining the total mass of the raw material, the effective purity of the metal ion compound, the mass fraction of the metal element in the compound and the non-negative constraint, the actual doping amount is calculated, and the feeding mechanism is controlled to release the metal dopant. The raw materials with added metal dopants are sequentially subjected to carbon-mixed grinding, spray drying molding, and high-temperature sintering solid-phase reaction in an inert atmosphere. The sintered materials are then broken up, crushed, and demagnetized to obtain lithium iron phosphate finished products.

[0005] This invention divides spectral data into a main peak core region and an interpeak transition region, and calculates feature point weights based on the signal amplitude and the absolute value of the second-order difference. This suppresses redundant information and noise interference, while preserving the core peak shape and detailed variation characteristics of lithium iron phosphate raw materials. By unifying the wavenumber sampling point index and setting two types of alignment windows, the same row of features at different sampling positions corresponds to consistent spectral positions and feature meanings, ensuring the interpretability of covariance and cross-covariance calculations. Furthermore, through internal correlation matrices, cross-domain correlation matrices, eigenvalue decomposition, and orthogonal projection, high-dimensional spectral features are mapped to low-dimensional raw material state coordinates, representing the correlation changes between the main peak core features and interpeak transition features. By using a preset interpeak feature direction weight vector for row aggregation, information loss caused by positive and negative correlation cancellation can be reduced; simultaneously, combined with the row mean offset vector, the absolute amplitude deviation information of the current batch spectrum is supplemented, so that the obtained L-dimensional current raw material state coordinates simultaneously contain correlation changes and absolute deviation features. Finally, the difference between the target state and the current state is mapped to the theoretical doping adjustment amount by using the pseudo-inverse matrix of the preset sensitivity matrix. Combined with the total mass of raw materials, the effective purity of metal ion compounds, the mass fraction of metal elements, and non-negative constraints, the actual doping amount is calculated to realize doping compensation control based on spectral state feedback, thereby improving the control accuracy and execution reliability of the metal doping process. Attached Figure Description

[0006] Figure 1 A schematic diagram of the cross-domain correlation matrix between the core features of the main peak and the transition features between peaks; Figure 2 This is a schematic diagram of an orthogonal basis obtained based on eigenvectors. Detailed Implementation

[0007] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0008] In the first embodiment, the present invention proposes an intelligent execution system for adaptive doping processes, comprising the following modules: The calculation module is used to acquire spectral data from K sampling locations of lithium iron phosphate raw materials to form a sample set; identify the main peak position according to the phase standard spectrum, and divide the single spectrum into the main peak core region and the inter-peak transition region using the unified wavenumber sampling point index as the horizontal axis reference; calculate the second-order difference value in each region, and take the local extreme point as the initial feature point; determine the weight according to the signal amplitude and the absolute value of the second-order difference, select the spectral amplitude feature value in the preset N main peak core alignment windows and M inter-peak transition alignment windows, and form the first and second feature vectors according to the fixed spectral position order.

[0009] Multi-point random scanning is performed on the lithium iron phosphate raw material to be doped on the production line. The number of sampling points is set to K, where K is an integer greater than or equal to 10. Each scan acquires raw spectral data containing the correspondence between spectral amplitude and wavenumber. The lithium iron phosphate raw material is the basic raw material or precursor material for preparing modified lithium iron phosphate, including at least a lithium source, an iron source, and a phosphorus source, and optionally also a carbon source, a dispersion medium, and process aids. The lithium source, iron source, and phosphorus source are prepared according to the stoichiometric ratio of lithium iron phosphate or a preset deviation ratio. The carbon source is at least one of sucrose, glucose, starch, citric acid, asphalt, PEG, DCDA, or fructose. If the original horizontal axis output by the detection equipment is wavelength, it is first converted to wavenumber coordinates according to the reciprocal conversion relationship between wavenumber and wavelength, and a unified spectral sampling point index is re-established according to the increasing direction of wavenumber. The identification of the main peak position, regional extension, feature point sorting, and feature matrix construction are all based on the wavenumber sampling point index. The raw spectral data acquired each time are subjected to baseline calibration to remove the influence of background noise drift, and a moving average filtering algorithm is used for smoothing to reduce high-frequency random noise. The K smoothed spectral data are numbered according to the sampling order to obtain a sample set for feature extraction.

[0010] The standard phase spectrum of an ideal lithium iron phosphate crystal is used. A peak-finding algorithm is employed to locate the peaks in the standard phase spectrum where the spectral amplitude exceeds a preset intensity threshold. The wavenumber sampling point index corresponding to each peak position is recorded as the position of all main peaks. For each single spectral data point in the sample set, a preset number of sampling points are extended towards both the decreasing and increasing directions of the wavenumber sampling point index, centered on the wavenumber sampling point index of each main peak position, to form interval boundaries. This wavenumber interval containing the main peak position is then extracted and designated as multiple main peak core regions.

[0011] When the sampling point spacing between two adjacent main peaks is less than the sum of the number of preset extended sampling points on both sides, the adjacent main peaks are first grouped into the same main peak cluster, or the extension length on the adjacent sides is shortened to ensure that the core regions of the adjacent main peaks do not overlap. When using the main peak cluster method, the coverage area between the outermost main peaks in the main peak cluster and the preset extension areas on both sides together constitute a merged main peak core region. The remaining wavenumber intervals in the single spectral data, excluding all main peak core regions, are extracted and set as multiple inter-peak transition regions in ascending order of wavenumber sampling point index. If there are no remaining undivided sampling points between two adjacent main peak core regions, no separate inter-peak transition region is generated for that adjacent position. Within all existing inter-peak transition regions, M inter-peak transition alignment windows are uniformly divided according to the fixed wavenumber sampling point index order. If the actual inter-peak transition regions are insufficient to form M alignment windows, the longer inter-peak transition regions are subdivided equally to make up for the number of windows; if there are no usable inter-peak transition regions in the entire spectrum, a preset inter-peak transition placeholder template is invoked. The preset inter-peak transition placeholder template includes M inter-peak reference numbers pre-set according to the phase standard spectrum; for each inter-peak reference number, the average smooth spectral amplitude of the corresponding reference sampling point or the corresponding reference interval in the average spectrum of the sample set is read and used as the first to the Mth inter-peak transition placeholder feature values, thereby keeping the second feature vector dimension fixed.

[0012] For the spectral data within the defined main peak core region, the central difference method is used to perform a second-order difference operation on the spectral amplitude of each discrete data point within the band, obtaining the second-order difference value of each data point within the main peak core region. The second-order difference value of each data point is compared with the second-order difference values ​​of its two adjacent data points. If the second-order difference value of a point is simultaneously greater than or simultaneously less than the values ​​of its two adjacent points, then that data point is determined to be a local extremum point and is used as the initial feature point of the main peak core region. The same central difference method and adjacent point comparison rules are used to process the spectral data within the defined inter-peak transition region, identifying all local extrema points as the initial feature points of the inter-peak transition region.

[0013] The spectral amplitude of each initial feature point in the original smoothed spectrum is extracted as the signal amplitude, and the absolute value of the corresponding second-order difference is calculated. Maximum and minimum value normalization is performed on the spectral amplitude and absolute value of the second-order difference of each initial feature point within the same region to obtain the signal amplitude normalization index and the second-order difference normalization index. These are then weighted and summed according to preset weighting coefficients to obtain the weight value of each initial feature point. Based on the phase standard spectrum or the average spectrum of the sample set, N main peak core alignment windows are pre-determined within the main peak core region, and M inter-peak transition alignment windows are pre-determined within the inter-peak transition region. Each alignment window is fixedly arranged according to the wavenumber sampling point index order to ensure that the same row of features at different sampling positions corresponds to the same or adjacent spectral positions.

[0014] In the main peak core region, for each main peak core alignment window, the initial feature point with the highest weight value within the window is selected. The spectral amplitude of this initial feature point in the smoothed spectrum is used as the output spectral amplitude feature value of that window. If no initial feature point exists within the window, the spectral amplitude of the center sampling point of the window in the smoothed spectrum is read as a substitute spectral amplitude feature value. The resulting N spectral amplitude feature values ​​are arranged in a fixed order according to the main peak core alignment window, forming a first feature vector of length N. In the inter-peak transition region, feature points are selected or substitute spectral amplitudes are read in the same way for each inter-peak transition alignment window. The resulting M spectral amplitude feature values ​​are arranged in a fixed order according to the inter-peak transition alignment window, forming a second feature vector of length M. For both the main peak core region and the inter-peak transition region, the wavenumber sampling point index of the initial feature point is used to determine the alignment window affiliation, and the absolute value of the second-order difference is used for weight sorting. Neither of these is used as a value written into the first or second feature matrix.

[0015] In an optional embodiment, the step of identifying the main peak position based on the phase standard spectrum and dividing a single spectrum into a main peak core region and an interpeak transition region using a unified wavenumber sampling point index as the horizontal axis reference includes: Extract the center wavenumber sampling point index corresponding to each main peak as the reference point; With each reference point as the center, a preset number of continuous sampling points are extended towards the wavenumber sampling point index in both the decreasing and increasing directions. When adjacent main peak core regions overlap, merging or truncation is performed to form each independent main peak core region. Extract the remaining undivided spectral sampling point interval between two adjacent main peak core regions and set it as an independent inter-peak transition region.

[0016] Assuming a total of 2000 spectral sampling points are obtained, and all original abscissas have been converted to wavenumber coordinates, the sampling point index is re-established according to the wavenumber increment direction. In actual execution, the key main absorption bands of lithium iron phosphate are identified through the phase standard spectrum. For example, if two main peaks are found to correspond to the 601st and 1101st sampling points respectively in the wavenumber increment indexing system, these two index values ​​are extracted as reference points.

[0017] Taking the 601st sampling point as an example, 30 consecutive sampling points are extended in both the decreasing and increasing directions of the wavenumber sampling point index, covering 61 consecutive sampling points from index 571 to index 631, thus obtaining the first independent main peak core region. Similarly, for the second reference point, 61 sampling points from index 1071 to 1131 are extracted as the second main peak core region. If the distance between two reference points is less than 60 sampling points, the two reference points are grouped into the same main peak cluster, or the extension length on their adjacent sides is shortened to ensure that the main peak core regions do not overlap.

[0018] After completing the division of the main peak core region, the remaining undivided spectral sampling point intervals between two adjacent main peak core regions are extracted. The 439 consecutive undivided sampling points between indices 632 and 1070 are uniformly divided into independent inter-peak transition regions. If there are no intermediate remaining sampling points after merging adjacent main peak core regions, no independent inter-peak transition region is generated at that adjacent position; in this case, M inter-peak transition alignment windows are divided from other existing inter-peak transition regions according to a fixed index order, or supplemented by equidistant subdivision of longer inter-peak transition regions when there are insufficient available inter-peak transition regions. If there are no available inter-peak transition regions in the current spectrum, a preset inter-peak transition placeholder template is called; when this template includes 3 inter-peak reference numbers, the average smoothed spectral amplitude of the average spectrum of the sample set at the 3 reference numbers or reference intervals is read respectively, and used as the 3 inter-peak transition placeholder feature values ​​to ensure that the second feature vector still has a fixed length. This segmentation method can isolate the key absorption peak and its surrounding fluctuation region, avoiding background noise interference.

[0019] In an optional embodiment, the second-order difference value is calculated in each region, and local extreme points are taken as initial feature points; weights are determined based on the signal amplitude and the absolute value of the second-order difference; spectral amplitude feature values ​​are selected within preset N main peak core alignment windows and M inter-peak transition alignment windows, and first and second feature vectors are formed according to a fixed spectral position order, including: Extract the signal amplitude of the i-th initial feature point and the absolute value of the second difference ; Normalize all initial feature points in the main peak core region or inter-peak transition region, and calculate the signal amplitude normalization index. The value is The difference after subtracting the minimum value of V1 is divided by the difference between the maximum value and the minimum value of V1. When the maximum value and the minimum value of V1 or V2 are equal, the corresponding normalization index is assigned a preset constant, preferably 0, to avoid the denominator being zero. and the second-order difference normalization index The value is The difference after subtracting the minimum value of V2 is divided by the difference between the maximum value and the minimum value of V2; According to the formula Calculate the final weight of the i-th initial feature point ; Based on the standard spectrum of the phase or the average spectrum of the sample set, N main peak core alignment windows are determined in the main peak core region, and M inter-peak transition alignment windows are determined in the inter-peak transition region. For each single spectrum, the initial feature point with the highest weight is selected in each main peak core alignment window and each inter-peak transition alignment window, and the spectral amplitude of the selected initial feature point in the smooth spectrum is used as the spectral amplitude feature value of the corresponding window; when there is no initial feature point in the corresponding alignment window, the spectral amplitude of the center sampling point of the alignment window in the smooth spectrum is read as the alternative spectral amplitude feature value. The selected spectral amplitude feature values ​​are arranged in order of fixed spectral position according to the alignment window, forming the first feature vector and the second feature vector respectively.

[0020] Suppose that five initial feature points are detected in a predefined main peak region, and the corresponding signal amplitude vector V1 and second-order difference absolute value vector V2 are recorded respectively. Assume the measured set of V1 is [1.2, 1.8, 2.5, 1.5, 2.0], with a maximum value of 2.5 and a minimum value of 1.2; and the measured set of V2 is [0.05, 0.12, 0.20, 0.08, 0.15], with a maximum value of 0.20 and a minimum value of 0.05.

[0021] For the third feature point out of the five initial feature points, the signal amplitude Equals 2.5, the absolute value of the second difference It equals 0.20. When calculating the normalization index at this point, substitute the actual value into the formula: Signal Amplitude Normalization Index. =(2.5-1.2) / (2.5-1.2)=1.0; Second-order difference normalization index =(0.20-0.05) / (0.20-0.05)=1.0. For the second feature point, the indices for that feature point are respectively... =(1.8-1.2) / (2.5-1.2)≈0.461, =(0.12-0.05) / (0.20-0.05)≈0.467.

[0022] After obtaining each indicator, substitute them into the weighting formula to calculate the final weight. The final weight of the third feature point. =0.7×1.0+0.3×1.0=1.0, and the final weight of the second feature point is... =0.7×0.461+0.3×0.467=0.4628. This normalization weighting process takes into account both the relative strength of the spectral peak amplitude and the degree of abrupt change in spectral curvature represented by the second-order difference. By assigning a fixed ratio of 0.7 to the signal amplitude and 0.3 to the second-order difference abrupt change, representative physicochemical characteristics can be retained in the feature point screening stage.

[0023] Continuing with the aforementioned main peak core region as an example, if the preset number of main peak core feature points N is 5, then according to the fixed index range of the main peak core region in the phase standard spectrum, the main peak core region is divided into 5 main peak core alignment windows according to the wavenumber sampling point index order. Within each main peak core alignment window, the final weights of the initial feature points are compared, and the initial feature point with the largest weight within that window is selected as the output spectral amplitude feature value for that window. If no second-order difference local extremum point is detected within a certain main peak core alignment window, the spectral amplitude of the center sampling point in the smoothed spectrum is directly read as the output spectral amplitude feature value for that window. Arranging the 5 main peak core alignment windows in order from low index to high index yields a first feature vector with a fixed length of 5.

[0024] For the interpeak transition region, if the preset number of interpeak transition feature points M is 3, the interpeak transition region is divided into 3 interpeak transition alignment windows in the same way, and the initial feature point with the largest weight is selected in each window; when there is no initial feature point in the window, the spectral amplitude of the sampling point at the center of the window in the smooth spectrum is used for filling. The 3 interpeak transition alignment windows are arranged in order from low index to high index to obtain a second feature vector with a fixed length of 3. Therefore, the first and second feature vectors obtained at different sampling positions both have a fixed length, and elements with the same index correspond to the same spectral alignment window.

[0025] Since the characteristic values ​​of each spectral amplitude are arranged in the order of the fixed wavenumber sampling points in the preset alignment window, the i-th element of the first feature vector obtained at different sampling positions corresponds to the same main peak core alignment window, and the j-th element of the second feature vector obtained at different sampling positions corresponds to the same inter-peak transition alignment window.

[0026] The module is used to traverse the sample set to generate aligned first and second feature matrices; calculate the row covariance of the first feature matrix to obtain the internal correlation matrix; calculate the cross-domain correlation matrix by calculating the cross-covariance of the two matrices; perform eigenvalue decomposition on the internal correlation matrix and extract the eigenvectors corresponding to the top L largest eigenvalues ​​to form an orthogonal basis; project the cross-domain correlation matrix onto the orthogonal basis to obtain the projected correlation matrix, and aggregate it according to the row weight vectors of the preset inter-peak feature directions. Combined with the row mean offset vectors of the two feature matrices, the L-dimensional raw material state coordinates are output.

[0027] Following the numbering order of the single spectral data within the sample set, traversing from the first to the Kth data, the above operations of region partitioning, second-order difference calculation, weight calculation, and feature point extraction are repeated. The K first feature vectors extracted from the first to the Kth spectral data are stacked side-by-side as column vectors in numerical order, forming a first feature matrix X with N rows and K columns. The i-th row of the first feature matrix X corresponds to the spectral amplitude feature sequence of the i-th main peak core alignment window at the K sampling positions. Similarly, the K second feature vectors extracted from the first to the Kth spectral data are stacked side-by-side in numerical order, forming a second feature matrix Z with M rows and K columns. The j-th row of the second feature matrix Z corresponds to the spectral amplitude feature sequence of the j-th inter-peak transition alignment window at the K sampling positions, completing the matrix representation of the entire sample set in the two-class region feature spaces.

[0028] After constructing the first feature matrix X and the second feature matrix Z, the row mean vector of the first feature matrix X and the row mean vector of the second feature matrix Z are retained. The row mean vector is used to characterize the absolute amplitude level of the current batch in the main peak kernel alignment window and the inter-peak transition alignment window. The de-meaning process is only used to calculate the correlation changes between features and does not delete the retained absolute amplitude information.

[0029] The first feature matrix X is subjected to row-wise mean reduction. The mean of each row of the first feature matrix X is calculated, and the mean of the corresponding row is subtracted from all elements in that row to obtain the mean-reduced centered first feature matrix. This centered first feature matrix is ​​multiplied by its transpose to obtain a product matrix with N rows and N columns. All elements of this product matrix are divided by (K-1) to obtain the unbiased estimated covariance matrix of the row dimension of the first feature matrix. This matrix is ​​set as the internal correlation matrix, which represents the correlation changes between the feature points of each main peak core under the same main peak core alignment system.

[0030] Similarly, the second feature matrix Z is subjected to row-wise mean reduction. The mean of each row of the second feature matrix Z is calculated, and each element is subtracted from the mean of the corresponding row to obtain the centered second feature matrix. The centered first feature matrix X is multiplied by the transpose of the centered second feature matrix Z to obtain a product matrix with N rows and M columns. All elements of this product matrix are uniformly divided by (K-1) to obtain the cross-covariance matrix between the features of the main peak core region and the features of the inter-peak transition region. This matrix is ​​designated as the cross-domain correlation matrix C, which is used to measure the degree of fusion influence between the feature information of different regions after spectral alignment.

[0031] Perform singular value decomposition or Jacobi diagonalization eigenvalue decomposition on the internal correlation matrix to obtain N real eigenvalues ​​and an eigenvector of length N corresponding to each eigenvalue. Sort the N eigenvalues ​​in descending order of numerical value. Calculate the ratio of the sum of the top L eigenvalues ​​to the sum of all N eigenvalues. Determine the number L to retain in dimensionality reduction by setting a threshold where the cumulative contribution rate exceeds a preset percentage. Extract the L eigenvectors corresponding to the L largest eigenvalues ​​and concatenate them column-wise to form a matrix with N rows and L columns. This matrix serves as the orthogonal basis for the mapping transformation.

[0032] The matrix constituting the orthogonal basis is transposed to obtain a transposed orthogonal basis matrix with L rows and N columns. This transposed orthogonal basis matrix is ​​then multiplied by a cross-domain correlation matrix C with N rows and M columns to generate a projected correlation matrix P with L rows and M columns. Based on qualified finished product samples or doped experimental samples, the inter-peak characteristic direction weight vector is pre-calibrated. The M elements in each row of the projected correlation matrix P are multiplied by their corresponding weight coefficients and then summed. This multi-column data is compressed and aggregated into a single column, resulting in a column vector containing L elements. This inter-peak characteristic direction weight vector is used to unify the contribution direction and intensity of different inter-peak transition features to the raw material state coordinates, reducing information masking caused by direct cancellation of positive and negative correlations. This column vector is used as the L-dimensional correlated state coordinates. Read the row mean vector of the first feature matrix X and the row mean vector of the second feature matrix Z, and concatenate them in a preset order to form the current row mean vector; subtract the current row mean vector from the standard row mean vector corresponding to the qualified finished product sample to obtain the absolute amplitude offset vector; map the absolute amplitude offset vector to L-dimensional absolute offset coordinates using a preset mean mapping matrix with L rows and the same number of columns as the dimension of the current row mean vector. Perform dimensionless processing on the L-dimensional associated state coordinates and the L-dimensional absolute offset coordinates according to preset standardization parameters; add the L-dimensional associated state coordinates and the L-dimensional absolute offset coordinates according to preset fusion weights to obtain the L-dimensional current raw material state coordinates, which simultaneously represent the associated change characteristics and absolute composition deviation characteristics of the current batch of raw materials to be doped.

[0033] In an optional embodiment, the step of calculating the row covariance of the first feature matrix to obtain the internal correlation matrix and calculating the cross-domain correlation matrix of the two matrices includes: The mean of each row of the first feature matrix is ​​calculated to form an N-dimensional vector. The mean of each row of the second feature matrix Z is calculated to form an M-dimensional vector. ; Based on the first feature matrix X, construct an N×N dimensional internal correlation matrix A, where the elements in the i-th row and r-th column of matrix A are... The calculation method is as follows: obtain and The first centralized difference is obtained. and The second centralization difference is obtained by multiplying the first centralization difference by the second centralization difference and summing the results over k from 1 to K, and then dividing the sum by (K-1). Based on the first feature matrix X and the second feature matrix Z, an N×M dimensional cross-domain correlation matrix C is constructed, where the element in the i-th row and j-th column of matrix C is... The calculation method is as follows: obtain and Find the first difference. and The second difference is obtained by multiplying the first difference by the second difference and summing the results over k from 1 to K, and then dividing the sum by (K-1). The resulting matrix A is used as the internal correlation matrix, and the resulting matrix C is used as the cross-domain correlation matrix.

[0034] Assume the sample set from the acquired lithium iron phosphate raw material contains K=11 independent spectral data samples. The number of main peak core spectral amplitude feature values, determined by a fixed alignment window order, is set to N=5, and the number of inter-peak transition spectral amplitude feature values ​​is set to M=3. In this case, the first feature matrix X has a dimension of 5×11, and the second feature matrix Z has a dimension of 3×11. The arithmetic mean is calculated for each row of data in matrix X; for example, averaging the values ​​of the 11 samples at the first main peak core feature position. Assume the mean is obtained... The mean is 1.5; similarly, calculate the mean of the first row of matrix Z. Assuming the result is 0.8, we obtain 5-dimensional vectors respectively. and 3D vector .

[0035] Calculate the first row and first column elements of the cross-domain association matrix C. At that time, the sample index k is traversed from 1 to 11. For the k-th sample, the element in the first row and k-th column of matrix X is calculated. First difference Find the element in the first row and k column of matrix Z. The second difference Multiply the two together. Summate the products obtained from the 11 iterations, assuming the sum is 5.0. After summing, divide the sum of 5.0 by the degrees of freedom (K-1), which is 10, to calculate... The value is 0.5. Following this logic, all row and column combinations of N=5 and M=3 are iterated to calculate all 15 cross-covariance elements. A 5×3 matrix C is output as the cross-domain correlation matrix, which measures the fusion fluctuation pattern of the main peak core characteristics and inter-peak transition characteristics of lithium iron phosphate at different sampling positions. The cross-domain correlation matrix between the main peak core characteristics and inter-peak transition characteristics is as follows: Figure 1 As shown.

[0036] In an optional embodiment, the eigenvalue decomposition of the internal correlation matrix, extracting the eigenvectors corresponding to the top L largest eigenvalues ​​to form an orthogonal basis, includes: Perform eigenvalue decomposition on the internal correlation matrix A and solve the characteristic equation. Obtain all eigenvalues and the corresponding feature vectors ; All eigenvalues Sort in descending order and select the L feature vectors corresponding to the top L largest feature values; Normalize the selected L mutually orthogonal eigenvectors to obtain orthogonal unit eigenvectors. The value is Divide by The L2 norm is formed by concatenating L orthogonal unit eigenvectors column-wise to form an orthogonal basis.

[0037] Setting the number of characteristic values ​​N for the main peak nuclear spectral amplitude to 5, the constructed internal correlation matrix A is a 5×5 real symmetric covariance matrix. Eigenvalue decomposition is then performed on matrix A, i.e., solving the system of equations. The algorithm calculates five real eigenvalues ​​and their corresponding five independent 5-dimensional eigenvectors. Assume the five calculated eigenvalues, arranged in descending order, are as follows: =12.5、 =3.2、 =1.1、 =0.4、 =0.1.

[0038] If the preset number of principal components L to be extracted is set to 3, then the top 3 largest eigenvalues ​​will be truncated, namely 12.5, 3.2, and 1.1, and their corresponding eigenvectors will be extracted. , , Assume the largest eigenvalue Corresponding feature vector The specific values ​​are [2.0, -1.0, 0.0, 1.5, -0.5]. Calculate the L2 norm of this vector, which is the square root of the sum of the squares of its elements, 2.738.

[0039] vector Dividing each element in the vector by its L2 norm of 2.738 to normalize it yields the standardized unit eigenvector. The values ​​are approximately [0.73, -0.36, 0.0, 0.54, -0.18]. Similarly, the normalized eigenvectors are calculated and obtained respectively. and By concatenating the three 5×1 unit vectors in parallel column order from left to right, a 5×3 orthogonal basis matrix is ​​generated as the final orthogonal basis. This operation utilizes principal component extraction to remove interference between high-dimensional feature data and suppress noise. The orthogonal basis vectors obtained through eigenvalue decomposition are as follows: Figure 2 As shown.

[0040] In an optional embodiment, projecting the cross-domain correlation matrix onto an orthogonal basis to obtain the projected correlation matrix includes: Obtain the transpose matrix of the orthogonal basis; Perform matrix multiplication on the transpose matrix and the cross-domain correlation matrix C to obtain matrix P; The output matrix P is used as the projection correlation matrix calibrated by the main peak structure mode.

[0041] Based on the aforementioned steps, a 5×3 orthogonal basis matrix is ​​generated, containing L=3 principal component vectors of N=5 dimensions. A 5×3 cross-domain correlation matrix C is also generated, where N=5 rows represent the core feature dimension and M=3 columns represent the inter-peak transition feature dimension. During the projection operation, the 5×3 orthogonal basis matrix is ​​transposed by interchanging rows and columns, resulting in a 3×5 transpose matrix. .

[0042] After obtaining the transpose matrix, perform matrix multiplication on it and the cross-domain association matrix C. Specifically, use the transpose matrix of dimension 3×5. Multiply by a 5×3 cross-domain association matrix C on the left. For example, when calculating the first row and first column element of the output matrix P, The first row vector, i.e. the transpose of the first principal component, and the first column vector of matrix C, i.e. the covariance of the first interpeak transition feature on each main peak feature, are summed by the element-wise product of their corresponding elements.

[0043] Following the matrix dimension matching rule, the multiplication operation (3×5)×(5×3) will output a 3×3 matrix P. This matrix P is then output as the projection correlation matrix calibrated by the main peak structure pattern. This spatial projection process maps and reduces the original cross-domain collaborative change information to the main feature orthogonal space of the main peak core structure.

[0044] In an optional embodiment, the output L-dimensional raw material state coordinates include: The projection correlation matrix P is an L×M dimensional matrix, and the dimension of the preset inter-peak feature direction weight vector is the same as the number of columns of the projection correlation matrix P. The projection correlation matrix P is aggregated by row direction. The row direction aggregation includes performing an inner product operation between the row vector corresponding to each row of the projection correlation matrix P and the preset inter-peak feature direction weight vector to obtain L row feature aggregation values. The preset inter-peak feature direction weight vector is pre-calibrated based on qualified finished product samples or doped experimental samples, and is used to unify the contribution direction and contribution intensity of different inter-peak transition features to the raw material state coordinates. The L row feature aggregation values ​​are arranged in row order to form a column vector with dimension L×1; Read the row mean vectors of the first feature matrix and the second feature matrix, and concatenate them in a preset order to form the current row mean vector; subtract the current row mean vector from the standard row mean vector of the qualified finished product sample to obtain the absolute amplitude offset vector; use a preset mean mapping matrix with L rows and the same number of columns as the current row mean vector to map the absolute amplitude offset vector into an L-dimensional absolute offset column vector; then, according to a preset fusion weight, weight and fuse the L-dimensional column vector composed of the row feature aggregation values ​​with the L-dimensional absolute offset column vector to obtain an L-dimensional current raw material state column vector; The merged L-dimensional column vector is output as the L-dimensional current raw material state coordinates representing the characteristics of the current raw material batch.

[0045] After obtaining the projected correlation matrix P consisting of L rows and M columns, assuming that the number of extracted principal components L equals 3 and the number of inter-peak transition feature points M also equals 3, the matrix presents as a 3×3 real number array. For this projected correlation matrix P, each row is weighted and aggregated according to a pre-defined inter-peak feature direction weight vector. Assume that the elements of the first row of the current projected correlation matrix P are [0.4, -0.1, 0.2], the elements of the second row are [0.1, 0.5, 0.0], and the elements of the third row are [-0.2, 0.1, 0.3], and assume that the preset inter-peak feature direction weight vector is [1, -1, 1].

[0046] For the first row of the matrix, calculate 0.4×1+(-0.1)×(-1)+0.2×1=0.7 as the first row feature aggregation value; using the same weighted aggregation logic, calculate 0.1×1+0.5×(-1)+0.0×1=-0.4 as the second value for the second row; calculate -0.2×1+0.1×(-1)+0.3×1=0.0 as the third value for the third row. Thus, a total of L=3 discrete row feature aggregation values ​​are calculated.

[0047] The three aggregated values ​​are arranged sequentially from top to bottom according to their row indices, constructing a 3×1 column vector. The specific values ​​of this column vector are represented as follows: This column vector is taken as an L-dimensional associated state column vector. The row mean vectors of the first feature matrix X and the second feature matrix Z are further read and concatenated in a preset order to form the current row mean vector. The difference between the current row mean vector and the standard row mean vector corresponding to the qualified finished product sample is calculated and mapped through a preset mean mapping matrix to obtain an L-dimensional absolute offset column vector. Then, according to a preset fusion weight, the L-dimensional associated state column vector and the L-dimensional absolute offset column vector are weighted and fused to obtain the final L-dimensional current raw material state column vector. This state variable is then fed into the doping compensation and control equation to participate in the calculation of the theoretical doping adjustment amount.

[0048] The execution module is used to multiply the pseudo-inverse matrix of the preset sensitivity matrix by the difference vector between the target state coordinates and the raw material state coordinates to obtain the theoretical doping adjustment amount; and combine the total mass of the raw material, the effective purity of the metal ion compound, the mass fraction of the metal element in the compound, and the non-negative constraint to calculate the actual doping amount and control the feeding mechanism to release the metal dopant.

[0049] Read the target state coordinates, which are L-dimensional column vectors calibrated based on historical data of qualified finished products. Subtract the element corresponding to the current raw material state coordinate from each element in the target state coordinates to obtain an L-dimensional difference vector representing the quality deviation. Obtain a preset sensitivity matrix that determines the influence of dopant element concentration changes on feature coordinates through orthogonal experiments. Solve the Mohr Penrose pseudo-inverse matrix of this preset sensitivity matrix using singular value decomposition or generalized inverse algorithm. Multiply the solved pseudo-inverse matrix from the left by the L-dimensional difference vector and perform a mapping operation to calculate the vector composed of the theoretical doping adjustment amount of each required dopant element. The dopant elements include at least one of Mg, Ti, V, Mn, and Nb, and the number of columns in the preset sensitivity matrix is ​​consistent with the number of dopant element types involved in the adjustment. Each column corresponds to the influence relationship of one dopant element on the L-dimensional raw material state coordinates. The preset sensitivity matrix has L rows and L columns, and its Mohr Penrose pseudo-inverse matrix has L rows and L columns, respectively.

[0050] Obtain the total mass data of the current batch of lithium iron phosphate raw materials to be processed on the production line. Each element in the calculated theoretical doping adjustment vector is considered as the mass fraction adjustment amount of the corresponding metal dopant element. This mass fraction adjustment amount is a dimensionless value, for example, 0.001 represents 0.1%. Multiply each element by the total weight of the raw materials to obtain the target compensation mass for each metal dopant element. For each metal dopant element, divide the target compensation mass of that metal element by the product of the effective purity of the corresponding metal ion compound and the mass fraction of that metal element in the compound to obtain the theoretical feed mass of the compound. Perform a non-negative constraint judgment on the theoretical feed mass. If the theoretical feed mass of a certain metal ion compound is less than 0, assign the feed mass of that compound to 0 to meet the actual process requirements of unidirectional addition; if it is greater than or equal to 0, leave it unchanged, thus establishing the final actual doping amount. Based on the actual doping amount, a drive control command is generated and sent to the servo motor and screw metering feeder at the bottom of the corresponding doping storage tank. This controls the feeding mechanism to release a metal ion compound containing at least one of Mg, Ti, V, Mn, and Nb according to the actual doping amount. The mixing and stirring equipment performs mixing operations according to preset process parameters. When the theoretical doping adjustment amount corresponding to a certain metal dopant element is negative and its absolute value exceeds the preset allowable deviation, a prompt message is output indicating that the batch needs to be diluted or returned to the upstream batching adjustment. Through this method, when the raw material state fluctuates, the amount of metal dopant added to the product can be automatically adjusted based on the spectral state coordinate deviation of the metal dopant elements requiring positive compensation, ensuring that the doped elements in the final product closely approximate the preset distribution requirements.

[0051] The finished product output module is used to sequentially process the raw materials after adding the metal dopant into carbon mixing and grinding, spray drying and molding, high-temperature sintering solid-phase reaction in an inert atmosphere, and to break up, crush and demagnetize the sintered materials to obtain lithium iron phosphate finished products.

[0052] Specifically, the raw materials, after initial mixing with the metal dopant, are transported to a sand mill. A carbon source and solvent are added simultaneously according to a preset formula ratio, and a dispersion grinding program is initiated for wet mixing. During the grinding process, the particle size is monitored in real time until the median particle size of the mixed slurry reaches a preset threshold, forming a homogeneous precursor slurry. The precursor slurry is then pumped to a spray dryer and dried under preset inlet and outlet air temperatures and atomization rate parameters. This allows the solvent in the slurry to evaporate rapidly, resulting in spherical or near-spherical precursor powders with specific morphology and bulk density. The precursor powder is loaded into a sintering kiln and subjected to high-temperature solid-state sintering under a continuously purged inert protective gas environment, such as nitrogen or argon, according to a set heating rate, high-temperature holding parameters, and cooling curve. In one embodiment, the sintering conditions are a temperature of 400-800°C, different heating times of 2-6 hours, and different holding times of 5-15 hours. During the sintering process, the raw material undergoes a phase change reaction to generate a lithium iron phosphate crystal structure. The added carbon source is heated and carbonized, forming a uniform conductive carbon coating layer on the surface of the crystal particles. The agglomerated material produced by sintering is then sequentially pulverized and graded using an airflow pulverizer and a grading and screening device to deagglomerate and disperse the material and filter out oversized particles. The pulverized and graded material is then subjected to multiple magnetic adsorption processes to remove free magnetic metal foreign matter. After passing inspection, the material is packaged to obtain an adaptively optimized lithium iron phosphate finished product. This lithium iron phosphate finished product can be used as a raw material for batteries to prepare lithium-ion battery cathode sheets.

[0053] In the second embodiment, the present invention also proposes an intelligent execution method for adaptive doping processes, comprising the following steps: A sample set is formed by acquiring spectral data from K sampling locations of lithium iron phosphate raw materials; the main peak position is identified according to the phase standard spectrum, and the single spectrum is divided into the main peak core region and the inter-peak transition region using the unified wavenumber sampling point index as the abscissa reference; the second-order difference value is calculated in each region, and the local extreme point is taken as the initial feature point; the weight is determined according to the signal amplitude and the absolute value of the second-order difference, and the spectral amplitude feature values ​​are selected in the preset N main peak core alignment windows and M inter-peak transition alignment windows, and the first and second feature vectors are formed according to the fixed spectral position order; The sample set is traversed to generate aligned first and second feature matrices; the row covariance of the first feature matrix is ​​calculated to obtain the internal correlation matrix; the cross-domain correlation matrix is ​​obtained by calculating the cross-covariance of the two matrices; the internal correlation matrix is ​​eigenvalued and the eigenvectors corresponding to the top L largest eigenvalues ​​are extracted to form an orthogonal basis; the cross-domain correlation matrix is ​​projected onto the orthogonal basis to obtain the projected correlation matrix, and the matrix is ​​aggregated according to the row direction of the preset inter-peak feature direction weight vector. Combined with the row mean offset vector of the two feature matrices, the L-dimensional raw material state coordinates are output. The theoretical doping adjustment amount is obtained by left-multiplying the pseudo-inverse matrix of the preset sensitivity matrix by the difference vector between the target state coordinate and the raw material state coordinate; and by combining the total mass of the raw material, the effective purity of the metal ion compound, the mass fraction of the metal element in the compound and the non-negative constraint, the actual doping amount is calculated, and the feeding mechanism is controlled to release the metal dopant. The raw materials with added metal dopants are sequentially subjected to carbon-mixed grinding, spray drying molding, and high-temperature sintering solid-phase reaction in an inert atmosphere. The sintered materials are then broken up, crushed, and demagnetized to obtain lithium iron phosphate finished products.

[0054] The above description represents the preferred embodiments of the present invention. It should be noted that, for those skilled in the art, various improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An intelligent execution system for adaptive doping process, characterized in that, include: The calculation module is used to acquire spectral data from K sampling locations of lithium iron phosphate raw materials to form a sample set; The main peak position is identified based on the standard phase spectrum, and the single spectrum is divided into the main peak core region and the inter-peak transition region using the unified wavenumber sampling point index as the horizontal axis reference. The second-order difference value is calculated in each region, and the local extreme point is taken as the initial feature point. The weight is determined based on the signal amplitude and the absolute value of the second-order difference. The spectral amplitude feature values ​​are selected in the preset N main peak core alignment windows and M inter-peak transition alignment windows, and the first and second feature vectors are formed according to the fixed spectral position order. The module is used to traverse the sample set to generate aligned first and second feature matrices; calculate the row covariance of the first feature matrix to obtain the internal correlation matrix; calculate the cross-domain correlation matrix by calculating the cross-covariance of the two matrices; perform eigenvalue decomposition on the internal correlation matrix and extract the eigenvectors corresponding to the top L largest eigenvalues ​​to form an orthogonal basis; project the cross-domain correlation matrix onto the orthogonal basis to obtain the projected correlation matrix, and aggregate it according to the row weight vector of the preset inter-peak feature direction; combine it with the row mean offset vector of the two feature matrices to output L-dimensional raw material state coordinates. The execution module is used to multiply the pseudo-inverse matrix of the preset sensitivity matrix by the difference vector between the target state coordinates and the raw material state coordinates to obtain the theoretical doping adjustment amount; and combine the total mass of the raw material, the effective purity of the metal ion compound, the mass fraction of the metal element in the compound and the non-negative constraint to calculate the actual doping amount and control the feeding mechanism to release the metal dopant. The finished product output module is used to sequentially process the raw materials after adding the metal dopant into carbon mixing and grinding, spray drying and molding, high-temperature sintering solid-phase reaction in an inert atmosphere, and to break up, crush and demagnetize the sintered materials to obtain lithium iron phosphate finished products.

2. The system of claim 1, wherein, The process of identifying the main peak position based on the standard phase spectrum and dividing a single spectrum into a main peak core region and an interpeak transition region using a unified wavenumber sampling point index as the horizontal axis reference includes: Extract the center wavenumber sampling point index corresponding to each main peak as the reference point; With each reference point as the center, a preset number of continuous sampling points are extended towards the wavenumber sampling point index in both the decreasing and increasing directions. When adjacent main peak core regions overlap, merging or truncation is performed to form each independent main peak core region. Extract the remaining undivided spectral sampling point interval between two adjacent main peak core regions and set it as an independent inter-peak transition region.

3. The system of claim 1, wherein, The process involves calculating second-order difference values ​​in each region, taking local extrema as initial feature points, determining weights based on signal amplitude and the absolute value of the second-order difference, selecting spectral amplitude feature values ​​within preset N main peak core alignment windows and M inter-peak transition alignment windows, and forming first and second feature vectors according to a fixed spectral position order, including: extracting a signal amplitude of the i-th initial feature point with the second order difference absolute value ; The normalization processing is respectively performed on all initial feature points in the main peak core region or the inter-peak transition region, and a signal amplitude normalization index is calculated When the maximum value of V1 or V2 is equal to the minimum value, the corresponding normalization index is assigned a preset constant; a second-order difference normalization index is calculated ; The weight of the i-th initial feature point is calculated according to the formula The weight of the i-th initial feature point is calculated according to the formula ; Based on the standard spectrum of the phase or the average spectrum of the sample set, N main peak core alignment windows are determined in the main peak core region, and M inter-peak transition alignment windows are determined in the inter-peak transition region. For each single spectrum, the initial feature point with the highest weight is selected in each main peak core alignment window and each inter-peak transition alignment window, and the spectral amplitude of the selected initial feature point in the smooth spectrum is used as the spectral amplitude feature value of the corresponding window; when there is no initial feature point in the corresponding alignment window, the spectral amplitude of the center sampling point of the alignment window in the smooth spectrum is read as the alternative spectral amplitude feature value. The selected spectral amplitude feature values ​​are arranged in order of fixed spectral position according to the alignment window, forming the first feature vector and the second feature vector respectively.

4. The system of claim 1, wherein, The calculation of the row covariance of the first feature matrix yields the internal correlation matrix; the calculation of the cross-domain correlation matrix by calculating the cross-covariance of the two matrices includes: computing the mean of each row of the first feature matrix to form an N-dimensional vector computing the mean of each row of the second feature matrix Z to form an M-dimensional vector ; Based on the first feature matrix X, construct an N×N dimensional internal correlation matrix A, where the elements in the i-th row and r-th column of matrix A are... The calculation method is as follows: obtain and The first centralized difference is obtained. and The second centralization difference is obtained by multiplying the first centralization difference by the second centralization difference and summing the results over k from 1 to K, and then dividing the sum by (K-1). Based on the first feature matrix X and the second feature matrix Z, an N×M dimensional cross-domain correlation matrix C is constructed, where the element in the i-th row and j-th column of matrix C is... The calculation method is as follows: obtain and Find the first difference. and The second difference is obtained by multiplying the first difference by the second difference and summing the results over k from 1 to K, and then dividing the sum by (K-1). The resulting matrix A is used as the internal correlation matrix, and the resulting matrix C is used as the cross-domain correlation matrix.

5. The system according to claim 1, characterized in that, The eigenvalue decomposition of the internal correlation matrix, extracting the eigenvectors corresponding to the top L largest eigenvalues ​​to form an orthogonal basis, includes: Perform eigenvalue decomposition on the internal correlation matrix A and solve the characteristic equation. Obtain all eigenvalues and the corresponding feature vectors ; All eigenvalues Sort in descending order and select the L feature vectors corresponding to the top L largest feature values; Normalize the selected L mutually orthogonal eigenvectors to obtain orthogonal unit eigenvectors. The value is Divide by The L2 norm is formed by concatenating L orthogonal unit eigenvectors column-wise to form an orthogonal basis.

6. The system according to claim 1, characterized in that, The step of projecting the cross-domain correlation matrix onto an orthogonal basis to obtain the projected correlation matrix includes: Obtain the transpose matrix of the orthogonal basis; Perform matrix multiplication on the transpose matrix and the cross-domain correlation matrix C to obtain matrix P; The output matrix P is used as the projection correlation matrix calibrated by the main peak structure mode.

7. The system according to claim 6, characterized in that, The output L-dimensional raw material state coordinates include: The projection correlation matrix P is an L×M dimensional matrix, and the dimension of the preset inter-peak feature direction weight vector is the same as the number of columns of the projection correlation matrix P. The projection correlation matrix P is aggregated by row direction. The row direction aggregation includes performing an inner product operation between the row vector corresponding to each row of the projection correlation matrix P and the preset inter-peak feature direction weight vector to obtain L row feature aggregation values. The preset inter-peak feature direction weight vector is pre-calibrated based on qualified finished product samples or doped experimental samples, and is used to unify the contribution direction and contribution intensity of different inter-peak transition features to the raw material state coordinates. The L aggregated row feature values ​​are arranged in row order to form a column vector with dimension L×1; Read the row mean vectors of the first feature matrix and the second feature matrix, and concatenate them in a preset order to form the current row mean vector; subtract the current row mean vector from the standard row mean vector of the qualified finished product sample to obtain the absolute amplitude offset vector; use a preset mean mapping matrix with L rows and the same number of columns as the current row mean vector to map the absolute amplitude offset vector into an L-dimensional absolute offset column vector; then, according to a preset fusion weight, weight and fuse the L-dimensional column vector composed of the row feature aggregation values ​​with the L-dimensional absolute offset column vector to obtain an L-dimensional current raw material state column vector; The merged L-dimensional column vector is output as the L-dimensional current raw material state coordinates representing the characteristics of the current raw material batch.

8. A smart execution method for an adaptive doping process, characterized in that, include: A sample set is formed by acquiring spectral data from K sampling locations of lithium iron phosphate raw materials; The main peak position is identified based on the standard phase spectrum, and the single spectrum is divided into the main peak core region and the inter-peak transition region using the unified wavenumber sampling point index as the horizontal axis reference. The second-order difference value is calculated in each region, and the local extreme point is taken as the initial feature point. The weight is determined based on the signal amplitude and the absolute value of the second-order difference. The spectral amplitude feature values ​​are selected in the preset N main peak core alignment windows and M inter-peak transition alignment windows, and the first and second feature vectors are formed according to the fixed spectral position order. The sample set is traversed to generate aligned first and second feature matrices; the row covariance of the first feature matrix is ​​calculated to obtain the internal correlation matrix; the cross-domain correlation matrix is ​​obtained by calculating the cross-covariance of the two matrices; the internal correlation matrix is ​​eigenvalued and the eigenvectors corresponding to the top L largest eigenvalues ​​are extracted to form an orthogonal basis; the cross-domain correlation matrix is ​​projected onto the orthogonal basis to obtain the projected correlation matrix, and the matrix is ​​aggregated according to the row direction of the preset inter-peak feature direction weight vector. Combined with the row mean offset vector of the two feature matrices, the L-dimensional raw material state coordinates are output. The theoretical doping adjustment amount is obtained by left-multiplying the pseudo-inverse matrix of the preset sensitivity matrix by the difference vector between the target state coordinate and the raw material state coordinate; and by combining the total mass of the raw material, the effective purity of the metal ion compound, the mass fraction of the metal element in the compound and the non-negative constraint, the actual doping amount is calculated, and the feeding mechanism is controlled to release the metal dopant. The raw materials with added metal dopants are sequentially subjected to carbon-mixed grinding, spray drying molding, and high-temperature sintering solid-phase reaction in an inert atmosphere. The sintered materials are then broken up, crushed, and demagnetized to obtain lithium iron phosphate finished products.

9. The method according to claim 8, characterized in that, The process of identifying the main peak position based on the standard phase spectrum and dividing a single spectrum into a main peak core region and an interpeak transition region using a unified wavenumber sampling point index as the horizontal axis reference includes: Extract the center wavenumber sampling point index corresponding to each main peak as the reference point; With each reference point as the center, a preset number of continuous sampling points are extended towards the wavenumber sampling point index in both the decreasing and increasing directions. When adjacent main peak core regions overlap, merging or truncation is performed to form each independent main peak core region. Extract the remaining undivided spectral sampling point interval between two adjacent main peak core regions and set it as an independent inter-peak transition region.

10. The method according to claim 8, characterized in that, The process involves calculating second-order difference values ​​in each region, taking local extrema as initial feature points, determining weights based on signal amplitude and the absolute value of the second-order difference, selecting spectral amplitude feature values ​​within preset N main peak core alignment windows and M inter-peak transition alignment windows, and forming first and second feature vectors according to a fixed spectral position order, including: Extract the signal amplitude of the i-th initial feature point and the absolute value of the second difference ; Normalize all initial feature points in the main peak core region or inter-peak transition region, and calculate the signal amplitude normalization index. The value is The difference after subtracting the minimum value of V1 is divided by the difference between the maximum value and the minimum value of V1. When the maximum value and the minimum value of V1 or V2 are equal, the corresponding normalization index is assigned a preset constant. and the second-order difference normalization index The value is The difference after subtracting the minimum value of V2 is divided by the difference between the maximum value and the minimum value of V2; According to the formula Calculate the final weight of the i-th initial feature point ; Based on the standard spectrum of the phase or the average spectrum of the sample set, N main peak core alignment windows are determined in the main peak core region, and M inter-peak transition alignment windows are determined in the inter-peak transition region. For each single spectrum, the initial feature point with the highest weight is selected in each main peak core alignment window and each inter-peak transition alignment window, and the spectral amplitude of the selected initial feature point in the smooth spectrum is used as the spectral amplitude feature value of the corresponding window; when there is no initial feature point in the corresponding alignment window, the spectral amplitude of the center sampling point of the alignment window in the smooth spectrum is read as the alternative spectral amplitude feature value. The selected spectral amplitude feature values ​​are arranged in order of fixed spectral position according to the alignment window, forming the first feature vector and the second feature vector respectively.

Citation Information

Patent Citations

  • Preparation method of nanometer lithium iron phosphate anode material for lithium ion battery

    CN103474661A

  • Lithium iron phosphate composite material and preparation method thereof, positive electrode sheet and battery

    CN118899425B