Battery material component intelligent identification method and system

Abstract

The invention relates to the technical field of component identification, in particular to an intelligent identification method and system for battery material components, and the method comprises the following steps: obtaining spectral data, calculating a local gradient change rate, screening background segments, generating a dynamic background, removing the background, extracting peak coordinates, generating an interval difference sequence, calculating an intensity fluctuation rate, and calling characteristic parameters in a classified manner. And defining an analysis boundary, extracting multiple peak value features to construct a description vector, calculating a vector distance clustering classification label, and matching a fingerprint database to output a component identification result. According to the method, peak topological structure vectors are constructed through multi-dimensional parameters in cleaning signals, structural quantitative expression of component peak shapes is achieved, component affiliation labels are established and mapped to a standard fingerprint database in combination with spatial distance calculation and clustering division means between the vectors, the capacity of capturing spectral signal micro-features is enhanced, and the method is suitable for being applied to the field of spectrum detection. And the identification precision and the discrimination stability under the multi-component coexistence situation in a complex material system are improved.

Description

A method and system for intelligent identification of battery material composition Technical Field

[0001] This invention relates to the field of component identification technology, and in particular to a method and system for intelligent identification of battery material components. Background Technology

[0002] The field of battery material composition identification technology involves technical solutions for detecting, analyzing, and identifying various chemical elements or compounds in battery materials. Specifically, it includes spectral analysis methods for elemental composition, image recognition methods for microstructural components, and intelligent analysis methods for material characterization data. This field is widely used in the production quality control, material performance evaluation, and R&D process optimization of new energy batteries such as lithium-ion batteries, sodium-ion batteries, and solid-state batteries. The accuracy of composition identification directly affects the performance stability and safety of battery materials. The identification process typically involves multiple steps, including acquiring detection spectra after material preparation, cleaning raw data, extracting target component features, and modeling identification rules. Furthermore, in practical industrial applications, it is necessary to balance processing efficiency, adaptability to diverse material systems, and improved automation levels.

[0003] Traditional intelligent identification methods for battery material composition refer to technical solutions that identify the composition of battery materials through manual analysis or simple rule-based judgment. These methods are mainly based on raw detection results such as inductively coupled plasma emission spectroscopy, X-ray diffraction pattern analysis, or scanning electron microscopy image analysis. By manually setting peak position comparison, diffraction angle position matching, or image morphology feature comparison, the specific chemical composition and crystal structure contained in the material are inferred, or classification is carried out by relying on empirical discrimination rules. The identification steps often rely on the intuitive judgment of the detection personnel on the original spectrum or image. The information processing flow lacks data-driven and feature learning mechanisms, making it difficult to cope with the identification challenges brought about by the complexity and diversity of material systems and spectrum changes. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method and system for intelligent identification of battery material components.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for intelligent identification of battery material composition, comprising the following steps:

[0006] S1: Obtain the original spectral data of the battery material, calculate the intensity difference to generate the local gradient change rate, filter the candidate segments of the background region according to the preset gradient change threshold, calculate the second-order difference, establish the background fitting base segment with stable sign, and construct the dynamic background curve.

[0007] S2: Based on the original spectral data of the battery material, remove the dynamic background curve, obtain the clean spectral signal, extract the peak position coordinates of the clean spectral signal, calculate the difference between adjacent peak position coordinates to generate an interval difference sequence, obtain the reference peak position set of the standard battery material spectrum, and compare the interval difference sequence with the reference peak position set to generate the peak position density fluctuation rate.

[0008] S3: Compare the peak density fluctuation rate with the preset battery system classification threshold to determine the system category to which the battery material belongs and obtain the feature extraction parameter set;

[0009] S4: Based on the feature extraction parameter set, define the target peak analysis boundary of the clean spectral signal, calculate the peak intensity, peak width at half maximum, descent slope ratio and symmetry ratio, and construct the peak topology description vector;

[0010] S5: Calculate the spatial distance between the peak topology description vectors, cluster the peak topology description vectors to establish a classification label set, map the classification label set to the standard component fingerprint library, and output the battery material component identification result.

[0011] The present invention improves upon this invention by including the following: the dynamic background curve includes a fitted node coordinate sequence, interpolation polynomial coefficients, and baseline intensity fitted values; the peak density volatility includes peak spacing dispersion, local density deviation values, and distribution uniformity indices; the feature extraction parameter set includes peak search window width, signal enhancement gain value, and edge detection sensitivity; the peak topology description vector includes vertex amplitude feature components, morphological width feature components, and contour asymmetry factors; and the battery material composition identification results include principal component category names, relative component content values, and matching confidence scores.

[0012] The present invention is improved in that the step of obtaining the dynamic background curve is specifically as follows:

[0013] S111: Obtain the original spectral data of the battery material, extract the light intensity value corresponding to each wavelength point and calculate the intensity difference between adjacent points, construct the local gradient change rate based on the intensity difference, compare the local gradient change rate with the preset gradient change threshold point by point, filter the wavelength index range where the gradient value is continuously lower than the gradient change threshold, and generate candidate segments of the background region.

[0014] S112: Call the candidate segments of the background region, perform second-order difference operation on the boundary wavelength points within the segments, obtain the curvature change value, monitor the polarity consistency of the mathematical symbols of the second-order difference values ​​in the continuous wavelength sequence, eliminate non-stationary intervals with abrupt inflection points based on the symbol polarity stability, retain wavelength intervals where the symbol state remains constant, and establish background fitting base segments.

[0015] S113: Based on the background fitting base segment, extract the center position coordinates of each segment interval as interpolation anchor points, use the piecewise cubic spline function to perform smooth connection operation on the empty spectral regions between interpolation anchor points, solve the background intensity estimate value corresponding to each wavelength position, and combine the background intensity estimate values ​​at the wavelength points in spectral order to construct a dynamic background curve.

[0016] The present invention is improved in that the step of obtaining the peak density volatility is specifically as follows:

[0017] S211: Obtain the dynamic background curve, combine it with the original spectral data of the battery material, perform point-by-point subtraction operation on the intensity values ​​of the two at the corresponding wavelength points, remove the baseline drift component and retain the residual signal component to generate a clean spectral signal.

[0018] S212: Perform first derivative zero-crossing point detection on the clean spectral signal, locate the horizontal axis coordinate position of the local maximum point, obtain the peak position coordinates, calculate the numerical difference between the coordinates of consecutive adjacent peak positions in the sequence, and generate an interval difference sequence.

[0019] S213: Call the interval difference sequence, combine it with the preset reference peak set, extract the standard spacing features in the reference set, and calculate the peak density fluctuation rate.

[0020] The present invention is improved in that the formula for calculating the peak density volatility is specifically as follows:

[0021] ;

[0022] in, This represents the volatility of peak density. Represents the total number of elements in the sequence. Representing the The value of the peak interval. Representing the One reference interval value, This represents the coverage deviation weighting coefficient.

[0023] The present invention is improved in that the step of obtaining the feature extraction parameter set is specifically as follows:

[0024] S311: Obtain the peak density fluctuation rate, call the preset battery system classification threshold, compare the peak density fluctuation rate with each boundary value in the battery system classification threshold to determine the numerical range in which the peak density fluctuation rate falls, and match the corresponding digital identifier from the preset coding table according to the numerical range to generate the system classification index code.

[0025] S312: Call the system classification index code as a retrieval key value and input it into the preset material system mapping relationship table. Traverse the key value pairs stored in the mapping relationship table, find the battery chemical system name that has a unique correspondence with the system classification index code, define the specific chemical component system to which the current battery material data belongs, and generate a battery system category label.

[0026] S313: For the battery system category label, perform an addressing query in the preset strategy library to locate the processing strategy associated with the battery system category label, read the predefined peak search window width, signal enhancement gain coefficient and edge detection sensitivity threshold in the processing strategy, encapsulate each read parameter, and generate a feature extraction parameter set.

[0027] The present invention improves upon the following: the process of setting the battery system classification threshold specifically involves collecting a standard spectral sample set covering multiple known battery material systems, calculating the corresponding sample peak density volatility for each sample set, statistically analyzing the probability density distribution characteristics of the volatility values ​​under each system category, defining the numerical critical point for distinguishing battery systems based on the intersection boundary points or confidence interval edges of multiple distribution curves, and establishing the critical point as the battery system classification threshold.

[0028] The present invention is improved in that the step of obtaining the peak topology description vector is specifically as follows:

[0029] S411: Based on the peak search window and edge detection threshold configured in the feature extraction parameter set, perform gradient scanning along the wavelength axis on the clean spectral signal, identify the starting inflection point of the signal amplitude jump and the ending inflection point of the fall back to the baseline, establish a wavelength index interval covering the complete peak shape and mark the effective data segments in the interval, and generate the target peak analysis boundary.

[0030] S412: Perform morphological feature operations on the spectral data segment within the analysis boundary of the target peak, extract the local maximum amplitude, calculate the spectral bandwidth at the 50% position of the peak height, calculate the quotient of the linear fitting slope of the falling edge and the rising edge, and solve the waveform integral area ratio on both sides of the peak center axis to obtain the peak intensity, peak width at half maximum width, falling slope ratio and symmetry ratio.

[0031] S413: Call the peak intensity, peak width at half maximum, descent slope ratio and symmetry ratio, perform normalization processing on each feature data, unify the numerical dimensions, and perform ordered concatenation and vectorization encapsulation of the processed feature components according to the preset feature space mapping rules to construct a digital sequence representing the high-dimensional geometric properties of the peak and generate a peak topology description vector.

[0032] The present invention is improved in that the steps for obtaining the battery material composition identification results are specifically as follows:

[0033] S511: Call the peak topology description vector, calculate the Euclidean distance between the vector and the initial cluster center point, classify the vector into the corresponding cluster space according to the distance minimization principle, perform iterative update operation of cluster center coordinates until the position parameters converge and stabilize, assign a unique classification identifier code to the clusters including similar topology feature vectors, and establish a classification label set;

[0034] S512: Extract the average feature vector of each cluster for the classification label set, call the preset standard component fingerprint library, perform cosine similarity comparison between the average feature vector and the standard material fingerprint data stored in the library, calculate the correlation confidence value between the features of the sample to be tested and the known standard components, and generate a component feature matching degree list.

[0035] S513: Based on the component feature matching degree list, perform numerical descending sorting, filter the matching item index with the highest confidence value, and query the corresponding battery material chemical component name and crystal structure parameters in the database according to the index. The obtained attribute information is structured, encapsulated and output to generate battery material component identification results.

[0036] A battery material composition intelligent identification system, the battery material composition intelligent identification system being used to implement the above-mentioned battery material composition intelligent identification method, the system comprising:

[0037] The background curve recognition module acquires the original spectral data of the battery material, calculates the intensity difference to generate the local gradient change rate, filters candidate segments of the background region based on the preset gradient change threshold, calculates the second-order difference, establishes a background fitting base segment with stable sign, and constructs a dynamic background curve.

[0038] The peak density analysis module, based on the original spectral data of the battery material, removes the dynamic background curve, obtains the clean spectral signal, extracts the peak position coordinates of the clean spectral signal, calculates the difference between the position coordinates of adjacent peaks to generate an interval difference sequence, and compares the interval difference sequence with the reference peak set to generate the peak density fluctuation rate.

[0039] The feature parameter acquisition module compares the peak density fluctuation rate with a preset battery system classification threshold to determine the system category to which the battery material belongs and obtains a feature extraction parameter set.

[0040] The topology organization module, based on the feature extraction parameter set, defines the target peak analysis boundary of the clean spectral signal, calculates the peak intensity, peak width at half maximum, descent slope ratio, and symmetry ratio, and constructs a peak topology description vector.

[0041] The component identification and confirmation module calculates the spatial distance between the peak topology description vectors, clusters the peak topology description vectors to establish a classification label set, maps the classification label set to the standard component fingerprint library, and outputs the battery material component identification result.

[0042] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0043] In this invention, a dynamic background curve is constructed by differential calculation and interval determination of gradient change features in the original spectral data. A volatility index is generated by combining the peak position interval sequence with the density comparison of the reference set, thus establishing a classification and discrimination criterion for battery material systems. In clean signals, a peak topology vector is constructed with multi-dimensional parameters, integrating intensity, half-width at half-maximum, slope ratio, and symmetry features, realizing a structured and quantitative expression of component peak shapes. Combining spatial distance calculation between vectors and clustering methods, component attribution labels are established and mapped to a standard fingerprint database to complete the component identification process. This eliminates the reliance on manual experience and fixed peak templates, strengthens the identification model's ability to capture micro-features of spectral signals, and improves the identification accuracy and discrimination stability in complex material systems with multiple coexisting components. Attached Figure Description

[0044] Figure 1 is a flowchart of the method of the present invention;

[0045] Figure 2 is a flowchart of the present invention for obtaining dynamic background curves;

[0046] Figure 3 is a flowchart of the present invention for obtaining the peak density volatility.

[0047] Figure 4 is a flowchart of the process for obtaining the feature extraction parameter set according to the present invention;

[0048] Figure 5 is a flowchart of the present invention for obtaining the peak topology description vector;

[0049] Figure 6 is a flowchart of the process for obtaining the battery material composition identification results according to the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0051] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0052] Please refer to Figure 1. This invention provides a technical solution: a method for intelligent identification of battery material components, comprising the following steps:

[0053] S1: Obtain the original spectral data of the battery material, calculate the intensity difference value sequence between adjacent wavelength points to generate the local gradient change rate, select the continuous wavelength interval with the local gradient change rate lower than the set value as the candidate segment of the background region based on the gradient change threshold, perform the second-order difference value calculation on the boundary wavelength points of the candidate segment of the background region, confirm the interval where the second-order difference sign is continuous and stable as the background fitting base segment, call the cubic spline interpolation function to fit the background fitting base segment, and construct the dynamic background curve;

[0054] S2: Subtract the dynamic background curve from the original spectral data of the battery material to obtain a clean spectral signal, extract the peak position coordinates of the clean spectral signal, calculate the interval difference sequence between consecutive adjacent peak position coordinates, and generate the peak density fluctuation rate based on the density comparison calculation result between the interval difference sequence and the preset reference peak position set.

[0055] S3: Compare the peak density volatility with the preset battery system classification threshold, determine the system category of the battery material based on the comparison result, and call the corresponding feature extraction parameter set from the preset strategy library according to the system category.

[0056] S4: Define the analysis boundary of each target peak of the clean spectral signal based on the feature extraction parameter set. Calculate the peak intensity, peak width at half maximum (FWHM), descent slope ratio, and symmetry ratio for each target peak. Combine the peak intensity, peak width at half maximum (FWHM), descent slope ratio, and symmetry ratio to construct a peak topology description vector.

[0057] S5: Calculate the vector space distance between peak topology description vectors, input the peak topology description vectors into the K-means clustering algorithm for clustering to establish a classification label set, map the classification label set to the standard component fingerprint database, and output the battery material component identification result.

[0058] The dynamic background curve includes the fitted node coordinate sequence, interpolation polynomial coefficients, and baseline strength fitted value. The peak density volatility includes peak spacing dispersion, local density deviation value, and distribution uniformity index. The feature extraction parameter set includes peak search window width, signal enhancement gain value, and edge detection sensitivity. The peak topology description vector includes vertex amplitude feature components, shape width feature components, and contour asymmetry factor. The battery material composition identification results include principal component category name, relative component content value, and matching confidence score.

[0059] Please refer to Figure 2. The specific steps for obtaining the dynamic background curve are as follows:

[0060] S111: Obtain the original spectral data of the battery material, extract the light intensity value corresponding to each wavelength point and calculate the intensity difference between adjacent points, construct a local gradient change rate reflecting the steepness of the waveform based on the intensity difference, compare the local gradient change rate with the preset gradient change threshold point by point, filter the wavelength index range where the gradient value is continuously lower than the gradient change threshold, and generate candidate segments of the background region.

[0061] The specific process of setting the gradient change threshold is as follows: statistically analyze the numerical distribution characteristics of the local gradient change rate sequence, calculate the average value and standard deviation of the gradient values ​​in the whole spectrum, and use the sum of the average value and the preset multiple standard deviation as the gradient change threshold.

[0062] A UV-Vis-NIR spectrophotometer was used to perform a full-spectrum scan of the battery material samples to be tested (such as lithium iron phosphate or ternary materials at different aging stages). The spectral response range was set to... to Spectral resolution better than The points time is set to To ensure the signal-to-noise ratio, the photon counts captured by the detector are converted into a digitized raw spectral data matrix via a USB 3.0 interface, denoted as... ,in Indicates wavelength. This represents light intensity. After the data is read into memory, a dark current subtraction operation is first performed, which involves subtracting the sensor's thermal noise substrate measured under no-light conditions from the original spectrum point by point. Subsequently, point-by-point differencing is performed on the preprocessed discrete spectral sequence to extract local gradient features. A traversal index is set. From 1 to ( (Total number of spectral points), calculate adjacent wavelength points. and The absolute value of the strength difference between them is expressed by the formula as follows: This operation generates a gradient sequence corresponding to the wavelength. This sequence visually reflects the steepness of the spectral waveform: in the characteristic peak region, The value increases significantly due to rapid signal changes; however, in the background or baseline region, The numerical values ​​are mainly controlled by random noise and remain at a low level. To adaptively define the boundary between a "flat background" and a "signal peak," the gradient sequence needs to be... Statistical analysis was performed. First, the arithmetic mean of the gradient values ​​across the entire spectrum was calculated. and standard deviation Based on the normal distribution In principle, set a threshold for gradient change. In the specific experiment of this embodiment, the preset multiplier is set to 3, that is... This setting ensures Random noise fluctuations are contained within a threshold, thus effectively distinguishing signals from noise. Finally, the program iterates through the gradient sequence. Perform sliding window detection. Set a minimum continuous length threshold. (i.e., 10 consecutive wavelength points). When a signal is detected from the index... arrive The continuous sequence satisfies (For all) )and At that time, in the wavelength range This segment is marked as an independent candidate segment for the background region. This process filters out single-point low-gradient misclassifications caused by random high-frequency noise, ensuring that the selected candidate segment truly represents the baseline portion of the spectrum.

[0063] S112: Call the candidate segments of the background region, perform second-order difference operation on the boundary wavelength points within the segment, obtain the curvature change value, monitor the polarity consistency of the mathematical symbols of the second-order difference values ​​in the continuous wavelength sequence, eliminate non-stationary intervals with abrupt inflection points based on the stability of symbol polarity, retain wavelength intervals where the symbol state remains constant, and establish background fitting base segments.

[0064] In-depth second-order differential analysis was performed on the data within each segment to remove potentially confounding broad peak edges or baseline bulges. For the spectral intensity sequence within any candidate segment... The second-order difference value (i.e., the discrete curvature approximation) is calculated using the central difference method. The calculation formula is as follows: This value This reflects the concavity and convexity of the spectral curve: This indicates that the curve is concave downwards (which may be a valley or a flat area). This indicates the curve is convex upwards (potentially a shoulder). To ensure the stability of the background fit, the algorithm needs to monitor the consistency of the sign polarity of the second-order difference values. Define the sign function. ,like Then take 1, if Then take -1. Within each candidate segment, examine the continuous wavelength sequence. Whether a sign flip occurs. If the sign frequently alternates within a certain interval (e.g., more than two flips within 5 data points), the region is determined to be a non-stationary interval dominated by high-frequency noise and is removed. If the sign changes from positive to negative or from negative to positive, and the trend continues for more than a preset inflection point judgment length (e.g., 5 points), a sudden inflection point (e.g., the edge of an overlapping peak) is determined, and the inflection point and its neighboring data need to be removed from the candidate segment. After the above screening, only wavelength intervals where the sign state remains constant (i.e., continuously positive or continuously with small fluctuations close to zero) and the length meets the interpolation requirements are retained. Specifically, for each retained interval, the variance of its second difference is calculated. ,like Below the preset stability threshold If the wavelength range is found to be true, then this range is confirmed as the true physical baseline. These rigorously validated wavelength ranges are established as background fitting base segments for subsequent baseline reconstruction.

[0065] S113: Based on the background fitting base segment, extract the center position coordinates of each segment interval as the interpolation anchor point, use the piecewise cubic spline function to perform smooth connection operation on the empty spectral region between the interpolation anchor points, solve the background intensity estimate value corresponding to each wavelength position, and combine the background intensity estimate values ​​at the wavelength points in spectral order to construct a dynamic background curve.

[0066] For each defined background fitting base segment, the geometric center point of the wavelength range of that segment is extracted as the interpolation anchor point. For example, if a base segment covers a wavelength range of... to Then extract The wavelength and corresponding average intensity value at that location are used as the anchor point coordinates. In this way, a series of discrete anchor point sets are obtained across the entire spectrum. These anchor points are connected using piecewise cubic spline functions. Unlike simple linear interpolation, cubic spline interpolation requires that each piecewise interval... Construct a cubic polynomial on Strict boundary conditions were imposed: not only were the function values ​​required to be continuous at the anchor points, but the first derivative (slope) and second derivative (curvature) were also required to remain continuous at the anchor points. This ensured that the generated baseline curve was smooth and continuously differentiable overall, conforming to the continuous gradual change characteristics of the physical spectral baseline. After solving for the coefficients of all piecewise polynomials... Then, each wavelength point of the original spectrum is... Substitute the corresponding spline function In the calculation, the estimated background intensity value at that location is obtained. All calculated background intensity values ​​are recombined in spectral wavelength order to construct a complete dynamic background curve covering the entire wavelength range. This curve can accurately fit the broadband fluorescence background or instrument baseline drift in the spectrum, while perfectly avoiding characteristic signal peak regions.

[0067] Please refer to Figure 3. The specific steps for obtaining the peak density volatility are as follows:

[0068] S211: Obtain the dynamic background curve, combine it with the original spectral data of the battery material, perform point-by-point subtraction on the intensity values ​​of the two at the corresponding wavelength points, remove the baseline drift component and retain the residual signal component to generate a clean spectral signal.

[0069] Obtain the constructed dynamic background curve and raw spectral data of battery materials Before performing the subtraction operation, a wavelength axis alignment check is first performed to ensure that the wavelength indices of the two sets of data strictly correspond. Then, a point-by-point subtraction operation is performed: This process effectively removes baseline drift components (caused by Rayleigh scattering, fluorescence interference, or thermal drift) superimposed on the true signal. Results This refers to the residual signal component, which theoretically contains only Raman scattering or diffraction peaks of the material and a small amount of random white noise. To prevent local negative values ​​from occurring during subtraction (which is physically meaningless), a non-negativity constraint is applied to the calculation result: if the calculation result at a certain point is less than zero, it is forced to be set to zero or a very small positive value (e.g., ...). The generated clean spectral signal serves as a pristine data source for subsequent feature extraction, significantly improving the signal-to-noise ratio (SNR) and peak position identification accuracy. Experimental data show that this processing step typically improves the SNR of the spectrum. This allows the faint characteristic peaks that were originally submerged in the sloping baseline to be clearly visible.

[0070] S212: Perform first derivative zero-crossing detection on the clean spectral signal, locate the horizontal coordinate of the local maximum point, obtain the peak position coordinate, calculate the numerical difference between the coordinates of consecutive adjacent peaks in the sequence, and generate an interval difference sequence.

[0071] For clean spectral signals The peaks were precisely located using the first-order derivative zero-crossing method. First, a Savitzky-Golay filter (window width 7, polynomial order 2) was used to slightly smooth the clean signal to suppress high-frequency noise interference with derivative calculations, resulting in a smoothed signal. Next, calculate the sequence of first derivatives. Program Scan For a sequence, find points that satisfy the following two conditions: 1) the derivative changes from positive to negative (i.e., a zero-crossing point); 2) the original signal strength corresponding to this zero-crossing point. The noise level is higher than a preset noise threshold (e.g., the maximum signal amplitude). These two conditions together pinpoint the local maximum point, i.e., the location of the peak. Recording the horizontal coordinates (wavelength positions) of all detected peaks forms a set of peak locations. ,in Subsequently, the numerical differences between the coordinates of consecutive adjacent wave crests in the sequence are calculated to generate an interval difference sequence. ,in ,and This spacing sequence reflects the characteristic distribution of lattice vibration modes or interplanar spacing, and is key fingerprint information for identifying the crystal structure of materials.

[0072] S213: Call the interval difference sequence, combine it with the preset reference peak set, extract the standard interval features from the reference set, and use the formula:

[0073] ;

[0074] Calculate and obtain the peak density volatility;

[0075] in, This represents the volatility of peak density. The total number of elements in the sequence is represented by the number of values ​​contained in the interval difference sequence. The sequence index variable is determined by iterating through the sequence elements. Representing the The peak interval values ​​are obtained by reading the data at the corresponding index positions in the interval difference sequence. Representing the The reference interval values ​​are obtained by querying the corresponding standard interval data in the preset reference peak set. The weighting coefficient representing the coverage deviation is obtained by reading the preset dimensionless parameter configuration. This represents the sum of all peak interval values, obtained by performing an accumulation operation on the interval difference sequence. This represents the sum of all reference interval values, obtained by performing an accumulation operation on the interval data in the preset reference peak set;

[0076] Load a pre-defined set of reference peak positions for a specific battery material system (such as layered oxides) from the database, and calculate its corresponding reference interval sequence. To quantify the structural deviation between the test sample and the standard reference, the standard spacing feature in the reference set is extracted, and the peak density volatility is calculated using the following formula:

[0077] ;

[0078] Explanation of the logic and operation of formulas:

[0079] This formula provides a comprehensive evaluation of crystal structure variations through two independent dimensions:

[0080] Part 1 (Square Root Items): This represents the "root mean square of local relative deviation". This term is used to capture lattice non-uniformity distortion. When local defects, stress concentrations, or phase transformations occur within the material, causing disproportionate changes in the spacing between certain crystal planes, this value will increase significantly.

[0081] Part Two (Absolute Value Term): This represents the "overall cumulative deviation". This item is used to capture the overall scaling of the cell parameters.

[0082] For example, thermal expansion caused by temperature changes or overall cell volume shrinkage caused by lithium-ion insertion / extraction will lead to a proportional change in the spacing of all peak positions. This term accurately reflects such macroscopic structural evolution. 3. Coefficient Coverage deviation weighting coefficient is used to balance the contribution of local distortion and overall scaling to the final score, ensuring that changes in a single dimension do not dominate the results.

[0083] Actual calculation example: In a test of the cathode material of an aged battery, the extracted peak interval sequence and the corresponding standard reference interval data are shown in Table 1.

[0084] Table 1. Peak Interval Data Collection Table:

[0085] index Measured peak spacing (nm) Standard reference interval (nm) Relative deviation term 110.210.00.000400215.315.00.00040038.18.00.000156412.412.00.00111159.910.00.000100 surface

[0086] Parameter setting: Total number of sequence elements (See Table 1). Coverage Deviation Weighting Coefficient : Set as This coefficient is derived from training based on historical data. to It can effectively cover the conventional aging mode within the range.

[0087] Calculation process:

[0088] First, calculate the first part (local distortion):

[0089] Summing the relative deviation term:

[0090] ;

[0091] Calculate the root mean square:

[0092] ;

[0093] Next, calculate the second part (global scaling):

[0094] Total measured intervals:

[0095] ;

[0096] Sum of reference intervals:

[0097] ;

[0098] Calculate the absolute value of the ratio difference:

[0099] ;

[0100] Weighted calculation: ;

[0101] Finally, by combining the two parts, we obtain the peak density volatility. : .

[0102] Calculation results (Right now This is not simply statistical variance, but a comprehensive physical quantity. This value indicates that the material under test has undergone approximately [a change / transformation] relative to the standard crystal structure. The overall lattice expansion (manifested by the global term) is accompanied by approximately The internal non-uniform lattice distortion (manifested by the root-mean-square term). Compared to the traditional method that only compares the positions of the main peaks, this method introduces... The parameters significantly improve the sensitivity of structural change detection, enabling the effective identification of microstructural degradation signs that appear in the early stages of battery aging but have not yet led to macroscopic fracture.

[0103] Please refer to Figure 4. The specific steps for obtaining the feature extraction parameter set are as follows:

[0104] S311: Obtain the peak density volatility, call the preset battery system classification threshold, compare the peak density volatility with each boundary value in the battery system classification threshold, determine the numerical range in which the peak density volatility falls, match the corresponding digital identifier from the preset coding table according to the numerical range, and generate the system classification index code.

[0105] The process of setting the battery system classification threshold is as follows: collect a standard spectral sample set covering multiple known battery material systems, calculate the corresponding sample peak density volatility for each sample set, statistically analyze the probability density distribution characteristics of volatility values ​​under each system category, define the numerical critical point that distinguishes battery systems based on the intersection boundary points or confidence interval edges of multiple distribution curves, and establish the critical point as the battery system classification threshold.

[0106] Peak density volatility The input is fed into a preset battery system classifier. The classifier operates based on a pre-established "volatility-system" mapping logic and calls a preset set of battery system classification thresholds. The specific comparison logic is as follows: If It is classified as a "highly crystalline layered oxide system" (such as newly prepared NCM materials) and assigned the classification index code "IDX_001". If It is classified as an "olivine-type phosphate system" (such as LFP materials) and assigned the classification index code "IDX_002". If It is classified as a "spinel-type or highly disordered system" (such as LMO or severely aged materials) and assigned the classification index code "IDX_003". This is based on the calculations obtained in step S213. Compare it with the threshold: Therefore, the value is determined to fall into the first interval, generating the corresponding system classification index code "IDX_001". The classification threshold was set through statistical analysis of 3000 standard spectral samples covering five mainstream commercial battery materials (NCM523, NCM811, LFP, LCO, LMO). The classification threshold was determined by fitting the values ​​of each system. The Gamma distribution curve of the value is used to select the lower confidence limit at the intersection of different distribution curves. The confidence level is used as a distinguishing threshold to ensure the robustness of the classification.

[0107] S312: Call the system classification index code as the retrieval key value input to the preset material system mapping table, traverse the key-value pairs stored in the mapping table, find the battery chemical system name that has a unique correspondence with the system classification index code, define the specific chemical component system to which the current battery material data belongs, and generate the battery system category label;

[0108] The generated system classification index code "IDX_001" is used as a key to retrieve data from a pre-defined material system mapping table (HashMap). This mapping table stores the correspondence between index codes and metadata of specific chemical systems. A search is performed, and the battery chemical system name corresponding to "IDX_001" is found to be "Nickel-Cobalt-Manganese Layered Oxide (NCM)". Simultaneously, the specific chemical component system to which the current battery material data belongs is defined, and a battery system category label containing rich semantic information is generated. For example, the generated label is in JSON format: {"System": "NCM", "Structure": "R-3m", "State": "Fresh"}. This label not only clarifies the material type but also implicitly contains its crystal space group information, providing precise guidance for subsequently invoking specific peak-fitting strategies.

[0109] S313: For the battery system category label, perform an addressing query in the preset strategy library to locate the processing strategy associated with the battery system category label, read the predefined peak search window width, signal enhancement gain coefficient and edge detection sensitivity threshold in the processing strategy, encapsulate each read parameter, and generate a feature extraction parameter set.

[0110] Extract the value "NCM" from the "System" field. Perform an address lookup in the preset strategy library. This strategy library stores optimized signal processing parameters for different material systems. For the "NCM" system, read the corresponding processing strategy: Peak search window width ( ): Set as This is because the characteristic peaks of NCM materials typically have a narrow half-width at half-maximum (FWHM), and a smaller window can avoid interference from adjacent peaks. Signal enhancement gain coefficient ( ): Set as Used to amplify weak Ni-O bond vibration signals in high wavenumber regions. Edge detection sensitivity threshold ( ): Set as Used to capture weak phase transition-related peaks. Each read parameter is encapsulated to generate a feature extraction parameter set. This avoids underfitting or overfitting problems caused by using a single fixed parameter to handle different materials. For example, if the label is LFP, a wider search window will be loaded (e.g., To accommodate its wider phosphate vibration band.

[0111] Please refer to Figure 5. The specific steps for obtaining the peak topology description vector are as follows:

[0112] S411: Based on the peak search window and edge detection threshold configured in the feature extraction parameter set, perform gradient scanning along the wavelength axis on the clean spectral signal, identify the starting inflection point of the signal amplitude jump and the ending inflection point of the fall back to the baseline, establish a wavelength index interval covering the complete peak shape and mark the effective data segments in the interval, and generate the target peak analysis boundary.

[0113] The process of performing gradient scanning along the wavelength axis on the clean spectrum signal is as follows: statistically analyzing the amplitude distribution dispersion of the clean spectrum signal, obtaining the global noise standard deviation, multiplying the global noise standard deviation with a preset signal-to-noise ratio gain coefficient to obtain the edge detection threshold, identifying all local maxima in the clean spectrum signal and calculating their average full width at half maximum (FWHM), setting a specified multiple of the average FWHM as the width limit of the peak search window; calculating the difference in intensity between adjacent wavelengths of the clean spectrum signal, constructing a first derivative sequence, selecting the first wavelength position in the first derivative sequence where the value is continuously positive and the cumulative amplitude increment exceeds the edge detection threshold, defining it as the starting inflection point, and selecting the wavelength position in the first derivative sequence where the value is continuously negative and the corresponding signal intensity returns to the zero baseline, defining it as the ending inflection point;

[0114] The process of establishing a wavelength index interval covering the complete peak shape and marking the valid data segments within the interval is as follows: pairing up the starting and ending inflection points that are adjacent on the time axis and meet the peak position search window width constraint; extracting all wavelength index values ​​between the paired starting and ending inflection points to construct a wavelength index interval; and extracting and marking the corresponding spectral intensity data sequences within the wavelength index interval as valid data segments.

[0115] Calculate the clean spectral signal in the peakless region (e.g.) The amplitude standard deviation of () is used as the global noise standard deviation. (measured as) (counts). This value is compared with the signal-to-noise ratio gain coefficient in the parameter set (here, ). Multiply by , and obtain the dynamic edge detection threshold. Counts. Then, construct the first derivative sequence. The algorithm scans along the positive wavelength axis. Find points that meet the following conditions: 1. Initial inflection point identification: Find points where the derivative values ​​of three consecutive wavelength points are positive, and the cumulative amplitude increment (i.e., the increase in signal value) exceeds The first position. Marked as 2. Termination Inflection Point Identification: After finding the starting point, continue scanning, looking for three consecutive wavelength points where the derivative value is negative and the corresponding signal intensity returns to near the zero baseline (e.g., less than 0.5%). The location of ) is marked as Pairing Combining and integrating the parameters in the set. Perform constraint verification (i.e., require) (Corresponding index number). If the verification passes, the interval is established as the target peak analysis boundary. This process accurately cuts out the data segment containing the complete peak shape from the full spectrum, removes background and neighboring peak interference, and generates "valid data segment".

[0116] S412: Perform morphological feature operations on the spectral data segment within the analysis boundary of the target peak, extract the local maximum amplitude, calculate the spectral bandwidth at the 50% position of the peak height, calculate the quotient of the linear fitting slope of the falling edge and the rising edge, and solve the waveform integral area ratio on both sides of the peak center axis to obtain the peak intensity, peak width at half maximum width, falling slope ratio and symmetry ratio.

[0117] Perform morphological feature operations to quantify the geometric properties of the peak. Peak intensity ( ): Directly search for the maximum value within the data segment. For example, Counts. Peak width at half maximum (FWHM) ): Find the intensity equal to the wave crest on both sides. (Right now The two wavelength positions (counts) and ,calculate For example, the calculation yields... nm. This is directly related to the crystallinity of the material; the better the crystallinity, the smaller the FWHM. (The following appears to be unrelated and possibly a separate sentence fragment: "decline ratio (") Linear fitting was performed on the data points along the rising edge to the left and the falling edge to the right of the peak to obtain the slope. and .calculate .For example, ,but The further this value deviates from 1, the more asymmetrical the peak shape. Symmetry ratio ( (): Using the wavelength corresponding to the peak as the central axis, divide the area under the peak into the left half. and the right half .calculate Ultimately, these four key physical characteristics are obtained as the basic data for describing the wave crest morphology.

[0118] S413: Call the peak intensity, peak width at half maximum, descent slope ratio and symmetry ratio, perform normalization processing on each feature data, unify the numerical dimensions, and perform ordered concatenation and vectorization of the processed feature components according to the preset feature space mapping rules to construct a digital sequence representing the high-dimensional geometric properties of the peak and generate a peak topology description vector.

[0119] The Min-Max normalization method is used to map each data item to a preset feature boundary value. Interval. For example: normalized intensity. Normalized half-width Normalized slope ratio Normalization symmetry Based on the preset feature space mapping rules, the processed feature components are concatenated sequentially to construct a four-dimensional vector. This vector is the peak topology description vector, which uniquely represents the geometric topological properties of the peak in high-dimensional space, providing a standardized mathematical object for subsequent clustering and identification.

[0120] Please refer to Figure 6. The specific steps for obtaining the battery material composition identification results are as follows:

[0121] S511: Call the peak topology description vector, calculate the Euclidean distance between the vector and the initial cluster center point, assign the vector to the corresponding cluster space according to the distance minimization principle, perform iterative update operation of cluster center coordinates until the position parameters converge and stabilize, assign a unique classification identifier code to the clusters including similar topology feature vectors, and establish a classification label set.

[0122] Peak topology description vector The input is fed into the dynamic clustering process based on the improved K-Means++ algorithm. The vector is then compared with the currently existing... Initialized cluster centers Euclidean distance between Assuming there are currently 3 cluster centers, the distances to each cluster are calculated as follows: Based on the principle of minimizing distance ( This vector is then assigned to the first cluster space. Subsequently, an iterative update operation is performed to update the cluster center coordinates. The new cluster center... It is recalculated from the average of all historical vectors and newly added vectors within the cluster. This process is repeated until the change in the center position parameter is less than the convergence threshold. At this point, the clustering is considered stable. A unique classification identifier (such as "Cluster_Alpha") is assigned to each cluster containing similar characteristics, and its attributes are stored in the classification label set. This step enables the automatic aggregation of massive amounts of unimodal data into statistically significant peak categories.

[0123] S512: Extract the average feature vector of each cluster for the classification label set, call the pre-set standard component fingerprint library, perform cosine similarity comparison between the average feature vector and the standard material fingerprint data stored in the library, calculate the association confidence value between the features of the test sample and the known standard components, and generate a component feature matching degree list.

[0124] Extract the average feature vector of each cluster It calls a pre-set standard component fingerprint library, which stores standard feature vectors of battery materials (such as LiNi0.5Co0.2Mn0.3O2) with known purity and proportions. Perform cosine similarity comparison, the calculation formula is:

[0125] ;

[0126] This calculation generates an association confidence score between 0 and 1. For example, the calculated similarity between the cluster to be tested and the "NCM523 standard sample" is... The similarity to the "NCM811 standard sample" is . These results are then compiled to generate a list of component feature matching degrees. Experiments show that the fingerprint matching method based on cosine similarity can achieve an accuracy of up to [percentage missing] in material composition identification. The above methods are significantly superior to traditional methods that rely solely on peak position matching.

[0127] S513: Based on the component feature matching degree list, sort the values ​​in descending order, filter the matching item index with the highest confidence value, and query the corresponding battery material chemical component name and crystal structure parameters in the database according to the index. The obtained attribute information is structured, encapsulated and output to generate battery material component identification results.

[0128] Filter out the values ​​with the highest confidence scores (e.g.) The index of matching items is used. Based on this index, a reverse query is performed in the background material property database. Detailed information corresponding to this index is retrieved, including: Chemical composition name: Li(Ni0.5Co0.2Mn0.3)O2; Crystal structure parameters: Unit cell parameters. Material state: early cycle life. Each attribute information obtained from the query is structured and encapsulated to generate a final battery material composition identification report, which is then output through a user interface or data interface. This result not only tells the user what the material is, but also provides in-depth information about its microstructure state, completing end-to-end intelligent analysis from the original spectrum to the material's physicochemical properties.

[0129] A battery material composition intelligent identification system is provided to implement the above-mentioned battery material composition intelligent identification method. The system includes:

[0130] The background curve recognition module acquires the original spectral data of the battery material, calculates the intensity difference to generate the local gradient change rate, filters candidate segments of the background region based on the preset gradient change threshold, calculates the second-order difference, establishes a background fitting base segment with stable sign, and constructs a dynamic background curve.

[0131] The peak density analysis module obtains the clean spectral signal by subtracting the dynamic background curve from the original spectral data of the battery material, extracts the peak position coordinates of the clean spectral signal, calculates the difference between the position coordinates of adjacent peaks to generate an interval difference sequence, and compares the interval difference sequence with the reference peak set to generate the peak density fluctuation rate.

[0132] The feature parameter acquisition module compares the peak density fluctuation rate with the preset battery system classification threshold to determine the system category to which the battery material belongs and obtains the feature extraction parameter set.

[0133] The topology organization module, based on the feature extraction parameter set, defines the target peak analysis boundary of the clean spectral signal, calculates the peak intensity, peak width at half maximum (FWHM), descent slope ratio, and symmetry ratio, and combines these parameters to construct a peak topology description vector.

[0134] The component identification and confirmation module calculates the spatial distance between peak topology description vectors, clusters the peak topology description vectors to establish a classification label set, maps the classification label set to the standard component fingerprint database, and outputs the battery material component identification results.

[0135] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for intelligent identification of battery material composition, characterized in that, Includes the following steps: S1: Acquire the original spectral data of the battery material, calculate the intensity difference to generate the local gradient change rate, filter candidate segments of the background region according to the preset gradient change threshold, calculate the second-order difference, establish the background fitting base segment with stable sign, and construct the dynamic background curve; S2: Based on the original spectral data of the battery material, remove the dynamic background curve, acquire the clean spectral signal, extract the peak position coordinates of the clean spectral signal, calculate the difference between adjacent peak position coordinates to generate the interval difference sequence, acquire the reference peak position set of the standard battery material spectrum, and compare the interval difference sequence with the reference peak position set to generate the peak position density fluctuation rate; S3: Compare the peak density fluctuation rate with the preset battery system classification threshold to determine the system category to which the battery material belongs and obtain the feature extraction parameter set; S4: Based on the feature extraction parameter set, define the target peak analysis boundary of the clean spectral signal, calculate the peak intensity, peak width at half maximum, descent slope ratio and symmetry ratio, and construct the peak topology description vector; S5: Calculate the spatial distance between the peak topology description vectors, cluster the peak topology description vectors to establish a classification label set, map the classification label set to the standard component fingerprint library, and output the battery material component identification result.

2. The intelligent identification method for battery material composition according to claim 1, characterized in that, The dynamic background curve includes a fitted node coordinate sequence, interpolation polynomial coefficients, and baseline strength fitting values. The peak density volatility includes peak spacing dispersion, local density deviation, and distribution uniformity index. The feature extraction parameter set includes peak search window width, signal enhancement gain, and edge detection sensitivity. The peak topology description vector includes vertex amplitude feature components, shape width feature components, and contour asymmetry factor. The battery material composition identification results include principal component category name, relative component content value, and matching confidence score.

3. The intelligent identification method for battery material composition according to claim 1, characterized in that, The specific steps for obtaining the dynamic background curve are as follows: S111: Obtain the original spectral data of the battery material, extract the light intensity value corresponding to each wavelength point and calculate the intensity difference between adjacent points, construct a local gradient change rate based on the intensity difference, compare the local gradient change rate with the preset gradient change threshold point by point, filter the wavelength index range where the gradient value is continuously lower than the gradient change threshold, and generate candidate segments of the background region; S112: Call the candidate segments of the background region, perform second-order difference operation on the boundary wavelength points within the segment, obtain the curvature change value, monitor the polarity consistency of the mathematical sign of the second-order difference value in the continuous wavelength sequence, eliminate non-stationary intervals with abrupt inflection points based on the stability of the sign polarity, retain the wavelength intervals where the sign state remains constant, and establish a background fitting base segment; S113: Extract the center position coordinates of each segment interval based on the background fitting base segment as interpolation anchor points, use the piecewise cubic spline function to perform smooth connection operation on the empty spectral regions between the interpolation anchor points, solve the background intensity estimation value corresponding to each wavelength position, combine the background intensity estimation values ​​at the wavelength points according to the spectral order, and construct a dynamic background curve.

4. The intelligent identification method for battery material composition according to claim 3, characterized in that, The specific steps for obtaining the peak density volatility are as follows: S211: Obtain the dynamic background curve, combine it with the original spectral data of the battery material, perform point-by-point subtraction on the intensity values ​​of the two at corresponding wavelength points, remove the baseline drift component and retain the residual signal component to generate a clean spectral signal; S212: Perform first derivative zero-crossing detection on the clean spectral signal, locate the horizontal axis coordinate position of the local maximum point, obtain the peak position coordinates, calculate the numerical difference between the position coordinates of consecutive adjacent peaks in the sequence, and generate an interval difference sequence; S213: Call the interval difference sequence, combine it with a preset reference peak set, extract the standard spacing features in the reference set, and calculate the peak density volatility.

5. The intelligent identification method for battery material composition according to claim 4, characterized in that, The specific formula for calculating the peak density volatility is as follows: ;in, This represents the volatility of peak density. Represents the total number of elements in the sequence. Representing the The value of the peak interval. Representing the One reference interval value, This represents the coverage deviation weighting coefficient.

6. The intelligent identification method for battery material composition according to claim 4, characterized in that, The specific steps for obtaining the feature extraction parameter set are as follows: S311: Obtain the peak density volatility, call the preset battery system classification threshold, compare the peak density volatility with each boundary value in the battery system classification threshold to determine the numerical range in which the peak density volatility falls, and match the corresponding digital identifier from the preset coding table according to the numerical range to generate a system classification index code; S312: Call the system classification index code as a retrieval key value and input it into the preset material system mapping relationship table, traverse the key-value pairs stored in the mapping relationship table, find the battery chemical system name that has a unique correspondence with the system classification index code, define the specific chemical component system to which the current battery material data belongs, and generate a battery system category label; S313: For the battery system category label, perform an addressing query in the preset strategy library to locate the processing strategy associated with the battery system category label, read the predefined peak search window width, signal enhancement gain coefficient, and edge detection sensitivity threshold in the processing strategy, encapsulate each read parameter, and generate a feature extraction parameter set.

7. The intelligent identification method for battery material composition according to claim 6, characterized in that, The process of setting the battery system classification threshold is as follows: a standard spectral sample set covering multiple known battery material systems is collected; the corresponding sample peak density volatility is calculated for each sample set; the probability density distribution characteristics of the volatility values ​​under each system category are statistically analyzed; based on the intersection boundary points or confidence interval edges of multiple distribution curves, the numerical critical point for distinguishing battery systems is defined, and the critical point is established as the battery system classification threshold.

8. The intelligent identification method for battery material composition according to claim 6, characterized in that, The steps for obtaining the peak topology description vector are as follows: S411: Based on the peak search window and edge detection threshold configured in the feature extraction parameter set, perform gradient scanning along the wavelength axis on the clean spectral signal, identify the starting inflection point of the signal amplitude jump and the ending inflection point of the fall back to the baseline, establish the wavelength index interval covering the complete peak shape and mark the effective data segments within the interval, and generate the target peak analysis boundary; S412: Perform morphological feature operation on the spectral data segments within the target peak analysis boundary, extract the local maximum amplitude, measure the spectral bandwidth at the 50% position of the peak height, calculate the quotient of the linear fitting slope of the falling edge and the rising edge, and solve the waveform integral area ratio on both sides of the peak center axis to obtain the peak intensity, peak width at half height, falling slope ratio and symmetry ratio; S413: Call the peak intensity, peak width at half maximum, descent slope ratio and symmetry ratio, perform normalization processing on each feature data, unify the numerical dimensions, and perform ordered concatenation and vectorization encapsulation of the processed feature components according to the preset feature space mapping rules to construct a digital sequence representing the high-dimensional geometric properties of the peak and generate a peak topology description vector.

9. The intelligent identification method for battery material composition according to claim 8, characterized in that, The steps for obtaining the battery material composition identification results are as follows: S511: Call the peak topology description vector, calculate the Euclidean distance between the vector and the initial cluster center point, assign the vector to the corresponding cluster space according to the distance minimization principle, perform iterative update operation of the cluster center coordinates until the position parameters converge and stabilize, assign a unique classification identifier code to the clusters including similar topology feature vectors, and establish a classification label set; S512: Extract the average feature vector of each cluster for the classification label set, call the preset standard component fingerprint library, perform cosine similarity comparison between the average feature vector and the standard material fingerprint data stored in the library, calculate the correlation confidence value between the features of the test sample and the known standard components, and generate a component feature matching degree list; S513: Perform numerical descending sorting based on the component feature matching degree list, filter the matching item index with the highest confidence value, query the corresponding battery material chemical component name and crystal structure parameters in the database according to the index, encapsulate and output each attribute information obtained from the query in a structured manner, and generate the battery material composition identification results.

10. A smart identification system for battery material composition, characterized in that, The system is used to implement the intelligent identification method for battery material components according to any one of claims 1-9. The system includes: a background curve identification module, which acquires the original spectral data of the battery material, calculates the intensity difference to generate a local gradient change rate, filters candidate segments of the background region according to a preset gradient change threshold, calculates the second-order difference, establishes a background fitting base segment with stable sign, and constructs a dynamic background curve; and a peak density analysis module, which, based on the original spectral data of the battery material, removes the dynamic background curve, acquires a clean spectral signal, extracts the peak position coordinates of the clean spectral signal, calculates the difference between adjacent peak position coordinates to generate an interval difference sequence, and compares the interval difference sequence with a reference peak position set to generate peak positions. The system comprises the following modules: a peak density volatility module and a feature parameter acquisition module. The former compares the peak density volatility with a preset battery system classification threshold to determine the system category to which the battery material belongs, and obtains a feature extraction parameter set. The latter, based on the feature extraction parameter set, defines the target peak analysis boundary of the clean spectral signal, calculates the peak intensity, peak width at half maximum (FWHM), descent slope ratio, and symmetry ratio, and constructs a peak topology description vector. The former, a component identification and confirmation module, calculates the spatial distance between the peak topology description vectors, clusters the peak topology description vectors to establish a classification label set, maps the classification label set to a standard component fingerprint database, and outputs the battery material component identification result.

Images