An impedance spectrum abnormal point automatic rejection method of an electrochemical workstation

Abstract

The application discloses an impedance spectrum abnormal point automatic elimination method of an electrochemical workstation, and particularly relates to the field of industrial data analysis, and is used for solving the problem that the sources of impedance spectrum abnormal points are difficult to distinguish. The impedance spectrum point data and collection quality information are acquired, and are associated as point-level spectrum point records according to the same frequency point index, so that the corresponding relationship between the spectrum shape data and the collection stability data is formed under the same frequency point. Then, the spectrum curve deviation, the collection stability deviation, the collection evidence state and the physical evidence state are generated based on the point-level spectrum point records, and the processing branches of each frequency point are output through a sequence state discrimination model. Subsequently, the test elimination, the weight reduction, the marking or the protection review are executed according to the processing branches, and the candidate processing spectrum is generated. Finally, the full-spectrum quality review is performed based on the candidate processing spectrum, the processing action is confirmed or the test elimination is cancelled according to the full-spectrum consistency change and the change of the key electrochemical parameters, and the processed impedance spectrum and the abnormal processing record are output.

Description

Technical Field

[0001] This invention relates to the field of industrial data analysis, and more specifically, to a method for automatically eliminating impedance spectrum anomalies in an electrochemical workstation. Background Technology

[0002] When performing impedance spectroscopy (ISS) tests using an electrochemical workstation, small perturbation signals are typically applied sequentially at a set of discrete frequencies, and the real part, imaginary part, and phase of the impedance of the tested system at each frequency are recorded. Since an impedance spectrum is not an isolated result of a single frequency point, but rather a spectral trajectory formed by multiple frequency points in frequency order, whether a particular frequency point is abnormal usually needs to be determined in conjunction with the continuous changes in its adjacent frequency points. In testing scenarios such as batteries, corrosion, coatings, and fuel cells, impedance spectroscopy is often used to determine interfacial reactions, charge transfer, diffusion processes, or film states. Bending, arcing, or low-frequency shifts in local frequency bands do not necessarily indicate acquisition errors; they may also correspond to changes in the electrochemical response of the tested system itself.

[0003] Impedance spectroscopy testing is also susceptible to the influence of the acquisition link status. During frequency scanning, electrochemical workstations may experience range switching, response amplitude fluctuations, unstable phase readings, and inconsistent results from repeated measurements at the same frequency point. These factors can cause a particular frequency point to deviate from the continuous trajectory formed by adjacent frequency points in the complex impedance plane. Such deviations have a similar appearance to changes in the actual electrochemical response, potentially manifesting as increased local bends, discontinuous arc segments, or discrete single points. If only the geometric deviation of the impedance spectral points is observed, it is difficult to determine whether the frequency point is an occasional disturbance during acquisition, an anomaly that needs to be eliminated, or a genuine mechanistic change that should be retained.

[0004] Current methods for handling impedance spectroscopy anomalies typically rely on full-spectrum fitting errors, local residuals, statistical outliers, or manual thresholding. When a frequency point is discontinuous with adjacent frequencies, a common approach is to directly delete that frequency point as an outlier or reduce its reliability during the fitting process by adjusting the residual magnitude. While these methods can handle obvious outliers, their judgment is largely based on the spectral deviation itself, failing to fully utilize the quality information simultaneously generated by the electrochemical workstation during acquisition, such as response stability, phase stability, range status, and repeatability bias. This leads to problems such as: genuine mechanistic changes may be mistakenly deleted, acquisition perturbation points may be retained, and improvements in a single local index may mask anomalous shifts in key electrochemical parameters. Therefore, automatic removal of impedance spectroscopy anomalies requires simultaneously organizing evidence of spectral changes and acquisition stability at the same frequency point, and determining differentiated processing paths based on the combined state of these two evidences in the frequency sequence.

[0005] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an automatic impedance spectrum anomaly removal method for an electrochemical workstation. This method acquires impedance spectrum point data and acquisition quality information, and associates them into point-level spectral record entries according to the same frequency index, establishing a correspondence between spectral data and stable acquisition data at the same frequency. Based on the point-level spectral record entries, it generates spectral curve deviation, acquisition stability deviation, acquisition evidence status, and physical evidence status, and outputs processing branches for each frequency point through a sequence state discrimination model. Subsequently, it performs trial removal, weight reduction, marking, or protection verification according to the processing branches, generating candidate processed spectra. Finally, it performs a full-spectrum quality verification based on the candidate processed spectra, confirming the processing action or reversing the removal based on changes in full-spectrum consistency and key electrochemical parameters, and outputting the processed impedance spectrum and anomaly processing record to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: An automatic method for eliminating impedance spectral anomalies in an electrochemical workstation includes: Step S1: Obtain impedance spectral point data and acquisition quality information generated during the impedance spectral test at the electrochemical workstation, and associate the impedance spectral point data and acquisition quality information into point-level spectral point records according to the same frequency point index; Step S2: Based on point-level spectral point records, generate spectral curve deviation to characterize local spectral shape changes and acquisition stability deviation to characterize acquisition stability changes. Combine the acquisition evidence state and physical evidence state, and output the processing branch for each frequency point through the sequence state discrimination model. Step S3: Based on the processing branch, perform the corresponding processing actions in trial elimination, weight reduction, marking and protection review on the corresponding frequency points, and synchronize the processing actions with the frequency point index to generate candidate processing spectrum; Step S4: Perform full-spectrum quality verification based on candidate processed spectra, and confirm the corresponding processing actions based on changes in full-spectrum consistency and key electrochemical parameters; cancel trial rejection if no confirmation is made, retain the weighting and labeling results, and output the processed impedance spectrum and abnormal processing records.

[0008] Furthermore, in step S1, the data processing server generates frequency indexes with the same test batch as the reading boundary, and connects the impedance spectrum data and acquisition quality information according to the frequency indexes; frequency points that do not match acquisition quality information are written with acquisition quality missing markers, and the matched records form a point-level spectrum record sequence according to the logarithm of the frequency.

[0009] Furthermore, in step S2, the data processing server forms a normalized trend of adjacent frequency bands based on the real part of impedance, the imaginary part of impedance, and the logarithmic frequency position of adjacent frequency points in the point-level spectral point record sequence, and obtains the local turning angle from the directional change between adjacent normalized trends; the data processing server uses the median level and median deviation of the local turning angle within the adjacent frequency point window as a local geometric reference to normalize the degree of local turning angle deviation of the current frequency point, obtain the spectral curve deviation, and writes the physical evidence status based on the comparison result of the spectral curve deviation and the spectral curve deviation discrimination boundary.

[0010] Furthermore, in step S2, the sampling stability deviation is determined by response stability, phase stability, repeatability deviation, and impedance magnitude. The data processing server obtains the normalized repeatability deviation based on the impedance magnitude, extracts the stable acquisition reference from the stable range frequency point, and writes the sampling stability deviation into the acquisition evidence status after comparing it with the stable acquisition reference.

[0011] Furthermore, in step S2, the sequence state discrimination model is implemented using a hidden Markov model. The hidden Markov model forms the observation sequence with spectral bias and sampling stability bias, and uses the retained, strong candidate anomalies, suspected sampling disturbances and suspected real mechanisms as hidden states, and processes the processing branch through the maximum probability hidden state path output.

[0012] Furthermore, in step S3, the data processing server generates a processing action index table based on the processing branches; the processing branches form a temporary removal state for frequency points with strong candidate anomalies, the processing branches form a corresponding state in the deweighting state and the marking state for frequency points suspected of having disturbances, the processing branches form a protection and verification state for frequency points suspected of having the true mechanism, and the processing branches enter the candidate processing spectrum for the retained frequency points according to the original data.

[0013] Furthermore, in step S3, the weighting state determines the weighting coefficient based on the degree of deviation of the sampling stability deviation from the sampling stability deviation discrimination boundary, and binds the weighting coefficient to the corresponding frequency point index; the marking state is generated when the range state is in the range change neighborhood, the sampling evidence state is insufficient sampling evidence, or there is a missing sampling quality mark, and carries an abnormal source prompt.

[0014] Furthermore, in step S4, the data processing server constructs the original verification spectrum based on the point-level spectral point record sequence, and constructs the candidate verification spectrum based on the candidate processing spectrum. In the candidate verification spectrum, the frequency points in the temporary elimination state do not participate in the full spectrum consistency calculation, the frequency points in the reduced weight state are included in the full spectrum consistency calculation according to the reduced weight coefficient, and the frequency points in the marked state and the protected verification state retain the original data.

[0015] Furthermore, in step S4, the full spectrum quality verification includes the calculation of full spectrum consistency changes; the data processing server uses the same residual calculation method to obtain the original full spectrum consistency residual and the candidate full spectrum consistency residual for the original verification spectrum and the candidate full spectrum consistency residual, and forms the full spectrum consistency improvement result based on the original full spectrum consistency residual and the candidate full spectrum consistency residual, and reads the contribution ratio of each participating frequency point using the residual calculation method.

[0016] Furthermore, in step S4, the full-spectrum quality verification also includes the determination of changes in key electrochemical parameters; the data processing server extracts key electrochemical parameters from the original verification spectrum and the candidate verification spectrum respectively, forms reasonable results or abnormal results of parameter changes, and generates anomaly processing records based on the full-spectrum consistency improvement results and parameter change determination results.

[0017] The technical effects and advantages of the automatic impedance spectrum anomaly removal method for an electrochemical workstation of the present invention are as follows: This invention associates impedance spectral data with acquisition quality information using the same frequency index, enabling the analysis of spectral changes and acquisition stability at each frequency point within the same data record. Compared to methods that delete frequencies based solely on local residuals or statistical outliers, this invention can simultaneously examine local spectral changes in the complex impedance plane and stability changes in the acquisition path, reducing the risk of conflating acquisition perturbations, actual electrochemical response changes, and strong candidate anomalies.

[0018] This invention uses spectral deviation, acquisition stability deviation, acquisition evidence status, and physical evidence status to form the basis for frequency point determination. The sequence state discrimination model outputs processing branches sequentially along the frequency points, allowing isolated mutations, suspected acquisition perturbations, and continuous spectral changes to enter different processing paths. This processing method reduces branch jumps caused by single-point threshold judgments, ensuring that the frequency point processing logic remains consistent with the impedance spectrum sequence structure.

[0019] This invention performs a full-spectrum quality check after generating candidate processing spectra, and combines changes in full-spectrum consistency and key electrochemical parameters to confirm or cancel elimination actions, ensuring that local frequency point processing is not determined solely by a single local indicator. The synchronously output anomaly processing record retains the frequency point index, processing branch, processing action, and check results, providing traceable data evidence for the impedance spectroscopy anomaly processing process. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating an automatic method for eliminating impedance spectrum anomalies in an electrochemical workstation according to the present invention. Figure 2 This is a schematic diagram of the point-level spectral point record generation and evidence quantity extraction of the present invention; Figure 3 This is a schematic diagram of the output processing branch of the sequence state discrimination model of the present invention; Figure 4 This is a schematic diagram illustrating the generation of candidate processing spectra and the full-spectrum quality verification of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 - Figure 4 This invention provides an automatic method for eliminating impedance spectrum anomalies in an electrochemical workstation, comprising: This invention focuses on processing impedance spectral point data and acquisition quality information generated during impedance spectral testing at an electrochemical workstation. First, the two are associated as point-level spectral point records according to the same frequency index, ensuring that frequency, real part of impedance, imaginary part of impedance, phase and response stability, phase stability, range status, and repeatability deviation remain corresponding at the same frequency. Then, based on the point-level spectral point records, spectral curve deviation (characterizing local spectral shape changes) and acquisition stability deviation (characterizing acquisition stability changes) are calculated, and combined with the acquisition evidence status and physical evidence status to form the basis for frequency point determination. Subsequently, a sequence state discrimination model outputs processing branches along the frequency point order, allowing strong candidate abnormal frequency points, suspected acquisition disturbance frequency points, suspected true mechanism frequency points, and reserved frequency points to enter their respective processing paths. Based on this, candidate processing spectra are formed, and the processing actions are confirmed or revoked through full-spectrum quality review, ultimately resulting in the processed impedance spectrum and abnormal processing records.

[0023] Step S1: Obtain impedance spectral point data and acquisition quality information generated during the impedance spectral test at the electrochemical workstation, and associate the impedance spectral point data and acquisition quality information into point-level spectral point records according to the same frequency point index.

[0024] In this embodiment, all steps are executed by a data processing server. Step S1 uses the impedance spectral point data and acquisition quality information output by the electrochemical workstation during an impedance spectral test as input, and establishes a one-to-one correspondence between the impedance spectral point data and the acquisition quality information according to the same frequency index, so that the resulting point-level spectral record can be used as the input object for step S2 to calculate the spectral deviation, acquisition stability deviation, and perform sequence state discrimination. The impedance spectral point data includes frequency, real part of impedance, imaginary part of impedance, and phase, and the acquisition quality information includes response stability, phase stability, range status, and repeatability deviation.

[0025] S101: Acquire impedance spectrum data and acquisition quality information.

[0026] The data processing server receives the test result file and acquisition record file corresponding to the same impedance spectroscopy test from the electrochemical workstation, and initiates step S1 with the same test batch as the reading boundary. The test result file reads the frequency, real part of impedance, imaginary part of impedance, and phase, while the acquisition record file reads the response stability, phase stability, range status, and repeatability deviation. Both retain the original frequency order of the electrochemical workstation's output.

[0027] During the reading process, the data processing server first generates a frequency index for each frequency point. The frequency index is a consecutive number generated according to the reading order within the same impedance spectroscopy test, used to maintain a one-to-one correspondence between impedance spectroscopy data and acquisition quality information in subsequent processing steps. If the electrochemical workstation has already output the frequency point number, the data processing server checks the consistency of that number with the reading order. If only the frequency sequence is output without the frequency point number, the data processing server generates a frequency index based on the position of the frequency record's row and stores the frequency as the subsequent sorting criterion under the same frequency index.

[0028] When the frequency unit, impedance unit, or phase unit in the test result file is provided by the electrochemical workstation, the data processing server reads it according to the unit identifier in the test result file. When the phase is expressed in degrees, the phase value involved in subsequent circle statistical calculations is first converted to radians for calculation. When the electrochemical workstation directly outputs response stability, phase stability, range status, and repeatability deviation, the data processing server directly reads the corresponding values. When the electrochemical workstation saves multiple response amplitude samples at the same frequency point but does not directly output the response stability, the data processing server reads all response amplitude samples at the same frequency point. The response stability is obtained by dividing the difference between the maximum and minimum response amplitude by the sum of the median of multiple response amplitude samples at that frequency point and the amplitude zero-prevention term. The amplitude zero-prevention term is taken as one-thousandth of the median of all valid response amplitude samples in the current test batch. The effective response amplitude sampling value is the response amplitude sampling value that can be read within the current test batch and is not marked as missing or invalid; when the electrochemical workstation saves multiple phase samples at the same frequency point but does not directly output the phase stability, the data processing server first converts the phase sampling values ​​into radians, then calculates the mean of the sine value and the mean of the cosine value of each phase sampling value, and uses the result of subtracting the square and square root of the two from 1 as the phase stability; when the electrochemical workstation only outputs the range switching frequency point, the data processing server records the range switching frequency point and the sequential position of its one adjacent frequency point as the range change neighborhood.

[0029] As an example, in battery electrochemical impedance spectroscopy detection, after the electrochemical workstation completes a scan from high frequency to low frequency, the test result file provides the frequency, real part of impedance, imaginary part of impedance, and phase for each frequency point. The acquisition and recording file provides the response stability, phase stability, range status, and repeatability deviation for the same frequency point. When the data processing server reads the first line of test results, it generates frequency point index 1 and simultaneously reads the corresponding quality information from the first line of the acquisition and recording file. When reading the second line of test results, it generates frequency point index 2 and continues to read the corresponding quality information until frequency point indices are obtained for all frequency points within a single impedance spectroscopy test.

[0030] S102: Establish the data correspondence under the same frequency point index.

[0031] After the initial reading is completed, the data processing server uses the frequency index obtained in S101 as the primary key to connect the impedance spectrum data with the acquisition quality information. During the connection process, the number of frequencies, frequency arrangement, and recording positions in the two data sources are compared first. When the number of frequencies is consistent and the frequencies at the same positions match, a one-to-one correspondence is directly established according to the frequency index. When the acquisition record file contains device frequency numbers or timestamps, the device frequency number is matched with the frequency index first, and then the frequency is used as the verification condition, ensuring that each piece of acquisition quality information corresponds to only one impedance spectrum data point.

[0032] When both the test result file and the acquisition record file contain frequency fields but their recording locations differ, the data processing server uses identical frequency values ​​or frequencies within the same device output accuracy range as matching criteria, connecting the acquisition quality information to the corresponding frequency index. The device output accuracy range is primarily determined by the frequency display accuracy, frequency unit identifier, or device exported accuracy field in the test result file or acquisition record file. When the file does not record the device output accuracy range, the data processing server uses one-thousandth of the minimum non-zero difference between adjacent frequency points within the same test batch as the frequency matching tolerance. The same acquisition quality information must not be connected to multiple frequency indexes. After matching, the data processing server uses the frequency, real part of impedance, imaginary part of impedance, and phase under each frequency index as an impedance spectrum point data group, and the response stability, phase stability, range status, and repeatability deviation under the same frequency index as an acquisition quality information group, generating point-level spectrum point records. Point-level spectral point records are data records formed with frequency point indices as the primary key. Their function is to carry within the same record the frequency, real part and imaginary part of impedance required for step S2 to calculate the spectral deviation, as well as the response stability, phase stability, repeatability deviation, and the input of the impedance modulus value determined by the real part and imaginary part of impedance.

[0033] If impedance spectrum data exists under the same frequency index but the acquisition quality information fails to match, the data processing server generates an acquisition quality missing marker under that frequency index. If acquisition quality information exists but impedance spectrum data fails to match, this acquisition quality information is not included in the point-level spectrum record sequence. The acquisition quality missing marker does not change the original values ​​of frequency, real part of impedance, imaginary part of impedance, and phase; it only serves as the input boundary when generating the acquisition evidence state in step S2, enabling subsequent processing to distinguish between data missingness and acquisition stability deviation.

[0034] S103: Form frequency point order and point-level spectral point record sequence.

[0035] After completing the correspondence of frequency point indices, the data processing server forms a frequency point order based on the frequencies in the point-level spectral point records. The frequency point order is arranged according to the logarithmic value of the frequency. The arrangement result is used in step S2 to calculate the spectral deviation between adjacent frequency points and is used by the sequence state discrimination model to output the processing branch along the frequency point sequence. The logarithmic frequency position of the i-th frequency point... By frequency Calculations show that ,in The frequency of the i-th frequency point is derived from the point-level spectral point record; The natural logarithm or a logarithm to base 10 can be used, and this should be kept consistent within the same impedance spectroscopy test. If If the value is not greater than 0, the frequency point will not participate in the frequency point order calculation and will be included in the marked point-level spectral point record set.

[0036] When two frequency points have the same frequency value, the data processing server maintains the order of the frequency point indices generated in step S1 as the sorting order, and retains the repeating frequency points in the point-level spectral point record sequence respectively. After the frequency point order is formed, the data processing server does not change the original values ​​in the point-level spectral point record, but only generates the mapping relationship between the frequency point order position and the frequency point index, so that the same frequency point maintains the same identity in the original record, geometric calculation, and acquisition stability calculation.

[0037] The data processing server then organizes the point-level spectral records into a point-level spectral record sequence. The point-level spectral record sequence is arranged according to frequency point order, and each sequence position points to a unique frequency point index. When reading the sequence in step S2, the local geometric input required for spectral deviation calculation can be directly generated by taking the real part of impedance, the imaginary part of impedance, and the logarithmic frequency interval corresponding to adjacent sequence positions. Alternatively, the acquisition stability input required for acquisition stability deviation calculation can be directly generated by taking the response stability, phase stability, and repeatability deviation under the same frequency point index.

[0038] For example, in impedance spectroscopy testing of corrosion samples, the original test results may be arranged in the output order of the electrochemical workstation. In S102, the data processing server has already connected the impedance spectral data and acquisition quality information of each frequency point into a point-level spectral record. After entering S103, the data processing server re-forms the frequency point order according to the logarithm of the frequency and saves the mapping from the frequency point index to the sequence position. When calculating the spectral deviation of the i-th non-first and last frequency point in the subsequent step S2, the real part of the impedance, the imaginary part of the impedance, and the frequency corresponding to the (i-1)-th, i-th, and i+1-th frequency points can be read simultaneously according to this mapping, without losing the original response stability, phase stability, range status, and repeatability deviation of that frequency point.

[0039] S104: Output the recorded results that can be called in step S2.

[0040] After the point-level spectral record sequence is formed, the data processing server performs a callability determination on each point-level spectral record. The determination criteria include whether the frequency index is unique, whether the frequency can participate in logarithmic sorting, whether the real and imaginary parts of the impedance can be read as complex impedance plane coordinates, response stability, phase stability, range status, and whether repeatability bias is connected to the same frequency index. Records that pass the determination are added to the valid point-level spectral record set, while records with missing acquisition quality markers or insufficient adjacent frequencies to form a local geometric input are added to the marked point-level spectral record set.

[0041] The effective point-level spectral point record set and the marked point-level spectral point record set together constitute the output of step S1. The effective point-level spectral point record set is used in step S2 to continuously calculate the spectral deviation and acquisition stability deviation; the marked point-level spectral point record set participates in the boundary determination of the acquisition evidence status or physical evidence status in step S2. The acquisition quality missing marker is used to indicate insufficient acquisition stability deviation input, and the boundary position of insufficient adjacent frequency points is used to indicate that the spectral deviation calculation needs to use a single-sided window or enter the physical evidence insufficient state. The data processing server also outputs the mapping relationship between the frequency point sequence position and the frequency point index, so that when steps S3 and S4 generate records for trial removal, weight reduction, marking, protection review, and anomaly handling, they can trace back to the original frequency point index formed in step S1.

[0042] After processing in step S1, the impedance spectral data and acquisition quality information output by the electrochemical workstation are associated as point-level spectral records with frequency index as the primary key. This further forms a point-level spectral record sequence, a set of valid point-level spectral records, a set of marked point-level spectral records, and a mapping relationship between the frequency order position and the frequency index. These results provide direct input for step S2 to generate acquisition evidence status, physical evidence status, spectral deviation, and acquisition stability deviation, and also preserve the traceable frequency source for the anomaly handling records output in step S4.

[0043] Step S2: Based on point-level spectral point records, generate spectral curve deviation to characterize local spectral shape changes and acquisition stability deviation to characterize acquisition stability changes. Combine the acquisition evidence state and physical evidence state, and output the processing branch for each frequency point through the sequence state discrimination model.

[0044] In this embodiment, step S2 is based on the point-level spectral record sequence, the set of valid point-level spectral records, the set of marked point-level spectral records, and the mapping relationship between the frequency point sequence position and the frequency point index formed in step S1. The point-level spectral records have completed the corresponding connection between impedance spectral data and acquisition quality information at the same frequency point. Step S2 calculates the spectral curve deviation and acquisition stability deviation in the frequency point sequence, generates the physical evidence state and the acquisition evidence state, and outputs the processing branch for each frequency point by the sequence state discrimination model.

[0045] S201: Calculate spectral deviation and generate physical evidence status based on point-level spectral record sequence.

[0046] The data processing server initiates processing from the point-level spectral record sequence output in step S1, reading the frequency, real part of impedance, and imaginary part of impedance one by one according to the frequency point sequence. The physical evidence state is a state result used to indicate whether a certain frequency point has sufficient complex impedance plane geometric evidence, including at least the spectral shape deviation physical evidence state, the spectral shape continuity physical evidence state, and the insufficient physical evidence state. It is jointly determined by the spectral shape deviation calculation result and the availability of adjacent frequency points, and is written back to the corresponding point-level spectral record along with the frequency point index.

[0047] For the i-th non-first and last frequency point, the data processing server first retrieves three point-level spectrum records with frequency point order positions of i-1, i, and i+1, which respectively form a complex impedance plane point composed of the real part and the imaginary part of the impedance. , and The complex impedance plane point at the i-th frequency. according to It means that, among them, Let be the real part of the impedance at the i-th frequency point. This represents the imaginary part of the impedance at the i-th frequency point, and both originate from the point-level spectral record corresponding to the same frequency index. The data processing server then uses the logarithmic value of the corresponding frequency to form the logarithmic frequency position. , and After ensuring that all three frequency points belong to the effective point-level spectral point record set and the adjacent logarithmic frequency intervals can be read, the data processing server calculates the normalized difference vectors on both sides of the i-th frequency point: ; in, The normalized difference vector on the left side of the i-th frequency point is derived from... to The complex impedance plane displacement and logarithmic frequency interval The normalized difference vector on the right side of the i-th frequency point is derived from... to The complex impedance plane displacement and logarithmic frequency interval; both are used to characterize the variation in the orientation of adjacent frequency bands in the complex impedance plane. or When the value is 0, the data processing server does not calculate the corresponding normalized difference vector and writes the frequency point into the state of insufficient physical evidence.

[0048] Subsequently, the data processing server according to and Calculate the local rotation angle using the cross product modulus and dot product. Local corners The change in spectral direction between two adjacent segments at the i-th frequency point: ; in, The local turning angle at the i-th frequency point is used to represent the degree of change in the impedance spectrum direction between two adjacent segments at that frequency point; This represents the magnitude of the cross product of two normalized difference vectors in the complex impedance plane. This represents the dot product of two normalized difference vectors. Used to output local rotation angles under the combined constraints of cross product modulus and dot product.

[0049] The adjacent frequency window consists of frequency points within the sequential positions of two frequency points before and after the i-th frequency point, where an effective local turn can be obtained. When the number of available frequency points is less than two, the adjacent frequency window is formed by using the available frequency points on one side of the i-th frequency point. If the available frequency points on one side still cannot form at least one effective local turn, the data processing server does not calculate the spectral deviation of that frequency point and writes the physical evidence status of that frequency point into the insufficient physical evidence state. The local geometric reference is the median level of the local turn within the adjacent frequency window. Spectral deviation of the i-th frequency point. Calculate using the following formula: ; in, The spectral deviation at the i-th frequency point is used to characterize the degree of deviation of this frequency point from the local bending level of the adjacent window; It is the set of effective local corners within the window of adjacent frequency points of the i-th frequency point; This represents the median level of the set. for Any local corner in; This represents the median deviation of the local rotation angle relative to the median level within an adjacent frequency window. To prevent tiny positive numbers with a denominator of 0, its value is taken as one-thousandth of the median level of all effective local turns in the current point-level spectral point record sequence.

[0050] At the first and last frequency points, the data processing server reads adjacent available frequency points according to the frequency point order position formed in step S1, and uses a single-sided window to form a local geometric reference. When the normalized difference vector lengths on both sides of the i-th frequency point are lower than the vector length discrimination boundary, or when the marked point-level spectral point record set indicates that the adjacent frequency points are insufficient to form a local geometric input, the data processing server does not output the effective spectral deviation of that frequency point and sets the physical evidence status to insufficient physical evidence. The vector length discrimination boundary is determined by the lower quartile of all computable normalized difference vector lengths in the current point-level spectral point record sequence after ascending sorting.

[0051] The boundary for judging the deviation of the score is denoted as The deviation of all valid spectrograms in the current point-level spectrogram record sequence. The upper quartiles are determined by sorting the values ​​in ascending order. At that time, the data processing server generated a spectral shape that deviated from the state of physical evidence; At that time, a continuous spectral shape physical evidence state is generated. As an example, in coating impedance spectrum detection, if a certain mid-frequency point forms a significant turn relative to the preceding and following frequency points in the complex impedance plane, the data processing server obtains a large local turn at that frequency point, writes the spectral deviation after robust normalization, and marks that frequency point as a spectral shape deviation physical evidence state.

[0052] S202: Calculate the acquisition stability deviation and generate the acquisition evidence status based on the acquisition quality information and impedance modulus.

[0053] After the spectral deviation and physical evidence status are established, the data processing server continues to use the mapping relationship between the sequential position of the same frequency point and the frequency point index to read the response stability, phase stability, repeatability bias, real part of impedance, imaginary part of impedance, and range status of each point-level spectral point record one by one. The acquisition evidence status is a result used to indicate whether a certain frequency point has evidence of acquisition link stability deviation, including at least the acquisition deviation acquisition evidence status, the acquisition stable acquisition evidence status, and the acquisition insufficient evidence status, which is jointly determined by the acquisition stability deviation, range status, and acquisition quality missing marker.

[0054] For the i-th frequency point, the data processing server first calculates the impedance magnitude from the real part and the imaginary part of the impedance. impedance magnitude Used to convert repeatability measurement bias into a comparison quantity that is independent of the impedance order of magnitude: ; in, Let be the impedance magnitude at the i-th frequency point. Let be the real part of the impedance at the i-th frequency point. is the imaginary part of the impedance at the i-th frequency point, and both are derived from the point-level spectral record corresponding to the frequency point index.

[0055] After the impedance magnitude is calculated, the data processing server divides the repetition error at the i-th frequency point by the impedance magnitude to obtain the normalized repetition error. : ; in, The normalized repeatability error at the i-th frequency point is used to eliminate the influence of different impedance levels on the comparison of repeatability measurement errors. The repetition measurement bias at the i-th frequency point originates from point-level spectral point records; To prevent tiny positive numbers with an impedance magnitude of 0 from rendering division impossible, the value is set to one-thousandth of the median level of the effective impedance magnitudes in the current point-level spectral record sequence. If the phase stability at the same frequency is given by the phase fluctuation output by the electrochemical workstation, the data processing server directly reads this phase stability. If the point-level spectral record contains multiple phase sampling results at the same frequency, the data processing server calculates the phase dispersion using circular statistics and uses the calculation result as the phase stability for subsequent comparisons.

[0056] The stable acquisition benchmark is extracted from the frequency points where the measurement range is stable. The data processing server first sorts the response stability and normalized repeatability deviation of the stable range frequency points in ascending order of their values. Since smaller values ​​for response stability and normalized repeatability deviation indicate lower acquisition fluctuations, the data processing server selects the frequency points within the top 50% of the samples in the ascending order as low-fluctuation candidate frequencies. The number of the top 50% is determined by rounding up half of the total number of samples. Only frequencies that simultaneously belong to the low-fluctuation candidate frequencies for both response stability and normalized repeatability deviation are included in the stable acquisition benchmark set. If the number of frequency points in the stable acquisition benchmark set is less than 10% of the total number of frequency points, the low-fluctuation candidate range is widened from the top 50% to the top 75%. If it is still less than 10% after widening, the stable acquisition benchmark is formed using all stable range frequency points. The aforementioned top 50% and top 75% are quantile ranges based on the data distribution of the current test batch, used to adaptively select relatively stable acquisition benchmarks under different sample impedance levels and different acquisition noise backgrounds.

[0057] The stable acquisition benchmark includes a response stability benchmark, a phase stability benchmark, and a normalized repetition error benchmark. The data processing server extracts the median levels of response stability, phase stability, and normalized repetition error from the corresponding frequency points of the stable acquisition benchmark to obtain the response stability benchmark. Phase stability benchmark and normalized repeatability bias benchmark For the i-th frequency point, the data processing server first generates the response stability ratio, phase stability ratio, and repetition error ratio. The response stability ratio is obtained by comparing the response stability of the i-th frequency point with a response stability benchmark; the phase stability ratio is obtained by comparing the phase stability of the i-th frequency point with a phase stability benchmark; and the repetition error ratio is obtained by comparing the normalized repetition error of the i-th frequency point with a normalized repetition error benchmark. Then, the sampling stability error is determined by combining the maximum ratio of these three ratios with their average ratio. Calculate using the following formula: ; in, Let be the sampling stability deviation at the i-th frequency point, used to characterize the degree of deviation of the acquisition process from the stable acquisition reference; The response stability at the i-th frequency point is derived from the point-level spectral point record; The phase stability at the i-th frequency point is derived from point-level spectral point records or circular statistical results of multiple phase samples at the same frequency point; To stabilize the median level of response stability in the baseline data acquisition; To stabilize the median level of phase stability in the acquisition reference; To stabilize the median level of normalized repeatability bias in the data acquisition baseline; To prevent tiny positive numbers with a denominator of 0, the value is set to one-thousandth of the median level of the corresponding benchmark quantity. The maximum ratio weight is set to a value between 0.5 and 0.8. The average ratio is weighted, with values ​​ranging from 0.2 to 0.5. The maximum ratio is used to retain the trigger sensitivity when any collection quality indicator deviates significantly, while the average ratio is used to reflect the cumulative effect when multiple collection quality indicators deviate slightly at the same time. As one implementation method, the data processing server uses the ratio of the difference between the maximum ratio and the average ratio to the maximum ratio as the degree of deviation concentration, and then... The value is determined to be 0.5 plus the product of the deviation from concentration and 0.3; when the calculated value is... If the value is less than 0.5, it is taken as 0.5. If the value is greater than 0.8, it is taken as 0.8.

[0058] The boundary for judging stability deviation is denoted as The sampling stability deviation at the corresponding frequency point of the stable sampling reference. Determined by the 90th percentile after sorting the values ​​in ascending order. At that time, the data processing server generates a state where the collected evidence deviates from the collected state; Furthermore, when the measurement range is stable, a stable acquisition evidence state is generated; when the i-th frequency point has a missing acquisition quality marker, an insufficient acquisition evidence state is generated. As an example, in fuel cell impedance spectrum detection, if the impedance spectrum data of a certain low-frequency point can still maintain continuity with adjacent frequency points, but the acquisition record shows that the range state is in the range change neighborhood and the repeated measurement deviation is higher than the impedance modulus, the data processing server writes the acquisition stability deviation of that frequency point into the corresponding point-level spectrum record and generates an acquisition deviation acquisition evidence state.

[0059] S203: Construct the observed sequence of the sequence state discrimination model and output the processing branch.

[0060] After obtaining the spectral deviation, sampling stability deviation, physical evidence status, and acquisition evidence status at each frequency point, the data processing server constructs the observation sequence of the sequence state discrimination model according to the frequency point order positions formed in step S1. The observation sequence is an ordered input sequence composed of the spectral deviation and sampling stability deviation of each frequency point, and the observation of the i-th frequency point is... ,in The spectral deviation at the i-th frequency point is... The sampling stability deviation for the i-th frequency point is derived from the calculation results of S201 and S202, respectively, and is written back to the same point-level spectrum record through the frequency point index. When there is insufficient physical evidence or insufficient sampling evidence for the i-th frequency point, the data processing server still retains... The sequence position; where there is insufficient physical evidence, the observation category classification is based solely on the sampling stability bias. Discrimination boundary with stability deviation The comparison results indicate that when there is insufficient evidence collected, the observation category classification is based solely on spectroscopic deviation. Discrimination boundary with score deviation The comparison results show that when both conditions are met, the observation category at that frequency point is classified into the first observation category.

[0061] In this embodiment, the sequence state discrimination model is implemented using a Hidden Markov Model (HMM). The HMM constrains the continuity of processing branches for adjacent frequency points through hidden state paths and distinguishes between isolated abrupt change frequency points and frequency points with continuous spectral changes. This HMM is a discrete observation probabilistic graphical model, and its model structure consists of a set of hidden states, a set of observation categories, an initial probability vector, a state transition matrix, and an observation probability matrix. The number of hidden states is four, corresponding to retained states, strong candidate anomalies, suspected acquisition perturbations, and suspected true mechanisms, respectively. The observations are two-dimensional vectors. The trainable parameters of this model are the initial probability vector, the state transition matrix, and the observation probability matrix.

[0062] The data processing server first determines the boundary based on the musical deviation. And the boundary of stability deviation judgment The observations at each frequency point are divided into four discrete observation categories: Category 1 observations are... and This indicates that both the spectral deviation and the sampling stability deviation are not higher than the corresponding discrimination boundary; the second category of observations is... and This indicates that the spectral deviation is higher than the spectral deviation discrimination boundary, while the sampling stability deviation is not higher than the sampling stability deviation discrimination boundary; the third category of observations is... and This indicates that the sampling stability deviation is higher than the sampling stability deviation discrimination boundary, while the spectral deviation is not higher than the spectral deviation discrimination boundary; the fourth category of observations is... and This indicates that both the spectral bias and the sampling stability bias are higher than the corresponding discrimination boundary. Therefore, the size of the observation probability matrix is... The size of the state transition matrix is The initial probability vector size is The observation probability matrix consists of rows corresponding to hidden states, arranged as: retained, strong candidate anomaly, suspected acquisition disturbance, and suspected true mechanism; and columns corresponding to observation categories 1 through 4. When no historical training data exists, the preset observation probability matrix is ​​as follows: the probabilities of the retained state corresponding to observation categories 1 through 4 are 0.85, 0.05, 0.05, and 0.05, respectively; the probabilities of the strong candidate anomaly state corresponding to observation categories 1 through 4 are 0.0667, 0.0667, 0.0667, and 0.80, respectively; the probabilities of the suspected acquisition disturbance state corresponding to observation categories 1 through 4 are 0.0833, 0.0833, 0.75, and 0.0833, respectively; and the probabilities of the suspected true mechanism state corresponding to observation categories 1 through 4 are 0.0833, 0.75, 0.0833, and 0.0833, respectively. When historical training data or intra-batch adaptive updates exist, the above-mentioned preset observation probability matrix is ​​replaced with the statistically normalized observation probability matrix.

[0063] For frequency points where both spectral deviation and sampling stability deviation are higher than the corresponding discrimination boundary, if the preceding and following frequency points do not belong to the same observation category, an initial state label for strong candidate anomalies is generated; for frequency points where sampling stability deviation is higher than the sampling stability deviation discrimination boundary but spectral deviation is not higher than the spectral deviation discrimination boundary, an initial state label for suspected acquisition disturbance is generated; for frequency points where spectral deviation is higher than the spectral deviation discrimination boundary but sampling stability deviation is not higher than the sampling stability deviation discrimination boundary, if the preceding or following frequency point also shows that spectral deviation is higher than the spectral deviation discrimination boundary but sampling stability deviation is not higher than the sampling stability deviation discrimination boundary, an initial state label for suspected true mechanism is generated; other frequency points generate retained initial state labels.

[0064] In one embodiment, the training dataset for the Hidden Markov Model (HMM) is formed from historical impedance spectrum test records. Each training sample includes spectral deviation, sampling stability deviation, physical evidence status, sampling evidence status, and processing branch labels obtained through manual verification under the same frequency index. The training dataset selects 200 impedance spectrum records, each containing 20 to 80 frequency points, with a total of no less than 4000 frequency points. Among these, the four types of processing branch labels—retention, strong candidate anomalies, suspected sampling disturbances, and suspected true mechanisms—each contain at least 300 frequency points. During training, the data processing server first counts the proportion of each processing branch label appearing at the first frequency point in frequency order as the initial probability, counts the number of transitions between processing branch labels at adjacent frequency points and normalizes them to obtain the state transition matrix, and counts the number of occurrences of the four discrete observation categories in each hidden state and normalizes them to obtain the observation probability matrix. When a certain count is 0, a smoothing count is added to the corresponding count, enabling the Viterbi algorithm to traverse all hidden state paths.

[0065] If no historical training data exists, the data processing server starts the sequence state discrimination model using preset initial parameters. The initial probability vector is set to... The vector contains four values ​​corresponding to retention, strong candidate anomaly, suspected acquisition perturbation, and suspected true mechanism, respectively. In the state transition matrix, the probability of transitioning from retention to retention is set to 0.80, from suspected true mechanism to suspected true mechanism to suspected true mechanism to suspected true mechanism to suspected true mechanism to suspected true mechanism to suspected true mechanism to 0.80, from strong candidate anomaly to retention to suspected true mechanism to suspected acquisition perturbation to suspected true mechanism ...

[0066] The data processing server determines the initial probabilities, observation probabilities, and state transition probabilities of the Hidden Markov Model (HMM) according to the priority of parameter sources. When a historical training dataset exists, the initial probability vector, state transition matrix, and observation probability matrix obtained from the historical training dataset are used preferentially. When no historical training dataset exists, the sequence state discrimination model is started using preset initial parameters. When intra-batch adaptive updates are needed using the current test batch, the data processing server calculates the number of times the state of the first frequency point, the number of state transitions at adjacent frequencies, and the number of times discrete observation categories appear in each hidden state based on the initial state markers generated from the current point-level spectral point record sequence. After adding a smoothing count, it re-normalizes to obtain the initial probabilities, state transition probabilities, and observation probabilities corresponding to the current test batch. After the parameters are determined, the data processing server inputs the observation sequence into the HMM and uses the Viterbi algorithm to output the maximum probability hidden state path.

[0067] When the Viterbi algorithm calculates the path score for each hidden state h at the i-th frequency, it first iterates through all possible hidden states g at the (i-1)-th frequency, calculates the product of the path score of the previous frequency and the state transition probability from g to h, and takes the maximum value. Then, it multiplies this maximum value by the observation probability of the current observation category output by hidden state h to obtain the maximum path score for the i-th frequency in hidden state h. After the data processing server completes the recursion sequentially along the frequency points, it backtracks to the maximum path and writes the hidden states in the maximum path into the processing branch of the corresponding frequency point.

[0068] For frequency points with insufficient physical evidence, the data processing server determines the acquisition direction observation only by whether the acquisition stability deviation is higher than the acquisition stability deviation discrimination boundary when classifying the observation category; for frequency points with insufficient acquisition evidence, the data processing server determines the physical direction observation only by whether the spectral deviation is higher than the spectral deviation discrimination boundary; if the same frequency point has both insufficient physical evidence and insufficient acquisition evidence, the observation category of the frequency point is classified into the category where the spectral deviation is not higher than the spectral deviation discrimination boundary and the acquisition stability deviation is not higher than the acquisition stability deviation discrimination boundary, and its final processing branch is determined by the state transition probability of adjacent frequency points.

[0069] As an example, in batch detection of battery impedance spectra, if a certain frequency point simultaneously exhibits a spectral deviation higher than the spectral deviation discrimination boundary and a sampling stability deviation higher than the sampling stability deviation discrimination boundary, but no identical deviations are observed at the preceding and following frequency points, the Viterbi algorithm classifies it as a strong candidate anomaly based on the frequency point order; if another range of adjacent frequency points continuously exhibits a spectral deviation higher than the spectral deviation discrimination boundary while the sampling stability deviation is not higher than the sampling stability deviation discrimination boundary, the data processing server classifies the continuous hidden state path as a suspected true mechanism; when several frequency points are in the range change neighborhood and the sampling stability deviation increases while the spectral deviation does not increase, the corresponding processing branch output is suspected acquisition disturbance.

[0070] S204: Synchronously output frequency point evidence results and processing branch results.

[0071] After the sequence state discrimination model outputs the processing branch for each frequency point, the data processing server uses the frequency point index as the primary key to synchronously write the spectral deviation, acquisition stability deviation, physical evidence status, acquisition evidence status, and processing branch into the point-level spectral point record, while maintaining the mapping relationship between the frequency point sequence position formed in step S1 and the frequency point index. After synchronous writing is completed, each frequency point generates a branch control result that can be directly read in step S3, as well as an evidence source result that can be used in step S4 to generate anomaly processing records.

[0072] The data processing server then groups the frequency indexes according to the processing branches, forming a set of strong candidate abnormal frequency points, a set of suspected acquisition disturbance frequency points, a set of suspected actual mechanism frequency points, and a set of reserved frequency points. Frequency indexes in the set of strong candidate abnormal frequency points are subject to trial elimination in step S3; frequency indexes in the set of suspected acquisition disturbance frequency points are subject to weight reduction or marking in step S3; frequency indexes in the set of suspected actual mechanism frequency points are subject to protection verification in step S3; and frequency indexes in the set of reserved frequency points are not processed in step S3. All of the above sets use the frequency index generated in step S1 as a unique reference key. When performing trial elimination, weight reduction, marking, or protection verification in step S3, the original frequency, real part of impedance, imaginary part of impedance, and phase can be read from the corresponding point-level spectral record, and spectral deviation, acquisition stability deviation, and processing branches are retained for writing into the abnormal processing record in step S4.

[0073] After processing in step S2, each frequency point in the point-level spectral point record sequence obtains spectral deviation, acquisition stability deviation, physical evidence status, acquisition evidence status, and processing branch, forming a set of strong candidate abnormal frequency points, a set of frequency points suspected of acquisition disturbance, a set of frequency points suspected of true mechanism, and a set of retained frequency points. The above results maintain traceable connections with frequency point indices, which can directly drive the trial elimination, weight reduction, marking, and protection review in step S3, and provide a basis for judgment for the output of abnormal processing records in step S4.

[0074] Step S3: Based on the processing branch, perform the corresponding processing actions in trial elimination, weight reduction, marking and protection review on the corresponding frequency points, and synchronize the processing actions with the frequency point index to generate candidate processing spectra.

[0075] In this embodiment, step S3 is based on the processing branches output from step S2, the set of strong candidate abnormal frequency points, the set of suspected acquisition disturbance frequency points, the set of suspected actual mechanism frequency points, the set of retained frequency points, and the point-level spectral records with spectral deviation and acquisition stability deviation already written in. Step S2 has completed the frequency point category determination. Step S3 generates trial elimination, weight reduction, marking, and protection review results according to different processing branches, and forms a candidate processing spectrum that can be handed over to step S4 for full-spectrum quality review.

[0076] S301: Frequency point processing action generated based on processing branch.

[0077] The data processing server initiates step S3 processing from the four frequency point sets formed in step S2, and uses the frequency point index as a unique reference key to read the frequency, real part of impedance, imaginary part of impedance, phase, spectral deviation, sampling stability deviation, acquisition evidence status, physical evidence status, and processing branch for each frequency point from the point-level spectrum record. After reading, the data processing server first establishes the correspondence between processing branches and processing actions according to the sequential position of the frequency points, so that each frequency point obtains a clear processing path before entering the candidate processing spectrum construction.

[0078] The processing action is a frequency point handling result marker triggered by the processing branch, used to indicate how the frequency point is retained in the candidate processing spectrum and how it participates in subsequent calculations. The data processing server maps the frequency point indices in the set of strong candidate abnormal frequency points to trial removal, maps the frequency point indices in the set of suspected disturbance frequency points to weight reduction or marking, maps the frequency point indices in the set of suspected true mechanism frequency points to protection verification, and maps the frequency point indices in the set of retained frequency points to retention. The above mapping is based on the processing branch output in step S2 and does not change the original values ​​of frequency, real part of impedance, imaginary part of impedance, and phase in the point-level spectrum point record.

[0079] When a suspected disturbance frequency point has a valid acquisition stability deviation and the acquisition evidence status is acquisition deviation, the data processing server maps the frequency point to a reduced-weight state. When a suspected disturbance frequency point is in the range variation neighborhood, has an acquisition quality missing marker, or the acquisition evidence status is insufficient acquisition evidence, the data processing server maps the frequency point to a marked state. If the same frequency point simultaneously meets the trigger conditions for both the reduced-weight state and the marked state, then the original data of that frequency point is retained in the candidate processing spectrum, along with a reduced-weight coefficient and an anomaly source indication.

[0080] After mapping is complete, the data processing server generates a processing action index table. This table is an intermediate result using frequency indexes as the primary key and processing branches and actions as callable content. It is used in subsequent steps S302, S303, and S304 to read the corresponding frequency points and construct candidate processing spectra. As an example, in batch detection of battery impedance spectra, if a frequency point is assigned to a set of strong candidate abnormal frequency points after step S2, the data processing server records this frequency point as a trial removal in the processing action index table. If another frequency point is assigned to a set of suspected acquisition disturbance frequency points after step S2, the data processing server records it as a downweighted or marked frequency. When multiple adjacent frequency points are assigned to a set of suspected true mechanism frequency points, the processing action index table continuously records the corresponding frequency points as protection verification.

[0081] S302: Perform trial elimination of strong candidate abnormal frequency points.

[0082] After the data processing server forms the action index table, it first reads the frequency point index for the action of trial removal and locates the corresponding frequency point from the point-level spectrum record sequence according to the frequency point's sequential position. The conditions for entering the trial removal process are: the frequency point index belongs to the set of strong candidate abnormal frequency points, and the corresponding processing branch is a strong candidate abnormality; the execution path is: temporarily remove the participation relationship of the frequency point in the candidate processing spectrum, while retaining the point-level spectrum record and its frequency point index.

[0083] Trial removal does not permanently delete the original impedance spectrum data points, but rather sets the corresponding frequency points to a temporary removal state during the candidate processing spectrum construction process. The temporary removal state indicates that the frequency point will not participate in the full-spectrum consistency calculation in step S4, but its participation can still be restored based on the full-spectrum quality review results. When the data processing server performs temporary removal for each strong candidate abnormal frequency point, it retains the frequency, real part of impedance, imaginary part of impedance, and phase of that frequency point in the point-level spectrum record, and removes the frequency point index from the frequency point participation list of the candidate processing spectrum, allowing step S4 to compare the changes in full-spectrum consistency before and after trial removal.

[0084] When multiple strong candidate abnormal frequency points appear adjacently, the data processing server writes them into a temporary rejection state one by one according to the frequency point order, and records adjacent test rejection frequency points as the same test rejection segment in the processing action index table. The test rejection segment is a segment result composed of frequency point indices that are all in the temporary rejection state in the consecutive frequency point order position, which is used in step S4 to restore the frequency point participation relationship by segment when canceling the test rejection. As an example, in coating impedance spectrum detection, if a certain mid-frequency point has both a spectral curve deviation higher than the spectral curve deviation discrimination boundary and a sampling stability deviation higher than the sampling stability deviation discrimination boundary, and shows an isolated abrupt change in adjacent frequency points, the data processing server writes the frequency point into a temporary rejection state. The candidate processing spectrum temporarily skips the frequency point in the subsequent consistency calculation, but the abnormal processing record can still read back its original impedance data and judgment basis according to the frequency point index.

[0085] S303: Reduce the weight or mark suspected disturbance frequencies.

[0086] After completing the trial elimination of strong candidate abnormal frequency points, the data processing server continues to read the frequency point indexes whose processing action is deweighting or marking, and reads the acquisition stability deviation, acquisition evidence status, and range status of the corresponding frequency points from the point-level spectrum record. The conditions for entering the deweighting or marking process are: the frequency point index belongs to the set of suspected acquisition disturbance frequency points, and the corresponding processing branch is suspected acquisition disturbance; the execution path is: based on whether the acquisition evidence status points to acquisition deviation and whether the range status is in the range change neighborhood, determine the contribution method and source indication method of the frequency point in the candidate processing spectrum.

[0087] For frequency points where a sampling stability deviation has formed and the sampling evidence status is sampling deviation, the data processing server retains the frequency, real part of impedance, imaginary part of impedance, and phase of that frequency point in the candidate processing spectrum and generates a weighted state. The weighted state indicates the reduced contribution of that frequency point in the full-spectrum consistency calculation or equivalent circuit fitting residual calculation in step S4. The weighting coefficient corresponding to the weighted state is determined according to the degree of deviation of the sampling stability deviation from the sampling stability deviation discrimination boundary. For the i-th frequency point, the data processing server sets the weighting coefficient. Its value satisfies And calculate using the following formula: ; in, is the weighting coefficient for the i-th frequency point, used to represent the contribution ratio of this frequency point in the full spectrum consistency calculation or the equivalent circuit fitting residual calculation; The sampling stability deviation at the i-th frequency point originates from step S2; The sampling stability deviation discrimination boundary is derived from the sampling stability deviation quantile results of the frequency point corresponding to the stable sampling reference in step S2. The higher the sampling stability deviation, The smaller the value, the lower the contribution ratio. When no frequency point generates a deweighted state, the data processing server will set the corresponding contribution ratio to 1. The candidate processing spectrum will be read during subsequent review. For the residual terms of the reduced-weighted frequency points, according to The frequency points to be included are counted at a contribution ratio of 1 for reserved frequency points, marked status frequency points, and protected review status frequency points.

[0088] For frequency points where the range status is in the range variation neighborhood, the acquisition evidence status is insufficient, or the acquisition quality is missing a marker, making it impossible to form a complete comparison of the acquisition stability deviation, the data processing server retains the frequency point in the candidate processing spectrum and generates a marker status. The marker status is used to indicate that the original impedance spectrum data of the frequency point is retained and an anomaly source prompt is added. The anomaly source prompt is bound to the frequency point index, and its content comes from the acquisition evidence status, range status, or acquisition quality missing marker. As an example, in fuel cell impedance spectrum detection, if the spectrum shape of a certain low-frequency point is still continuous with the adjacent frequency points, but the range status is recorded as the range variation neighborhood, the data processing server retains the frequency point in the candidate processing spectrum and writes it into the marker status; if the frequency point also has an acquisition stability deviation higher than the acquisition stability deviation discrimination boundary, the data processing server writes it into the deweighted status, so that when the frequency point is read for subsequent review, it can distinguish between the two processing results of retaining data and reducing contribution.

[0089] S304: Protect and verify suspected frequency points of the true mechanism and generate candidate processing spectra.

[0090] After trial removal, weight reduction, and marking are completed, the data processing server reads the frequency index of the processing action for protection review and identifies consecutive frequency points in the suspected true mechanism frequency point set based on the frequency point sequence position. The conditions for entering protection review are: the frequency point index belongs to the suspected true mechanism frequency point set, and the corresponding processing branch is suspected true mechanism; the execution path is: write the frequency point into the protection review status and maintain its eligibility to participate in the full spectrum quality review in the candidate processing spectrum.

[0091] The protection verification status indicates a processing state where electrochemical response information needs to be retained for a frequency point due to increased spectral deviation but where the sampling stability deviation does not exceed the sampling stability deviation discrimination boundary. The data processing server does not set a temporary removal status for suspected true mechanism frequencies. Instead, it writes the frequency index, spectral deviation, physical evidence status, and sequential position of consecutive frequencies into the processing action index table. This allows step S4 to combine changes in full-spectrum consistency and key electrochemical parameters during full-spectrum quality verification to determine whether a frequency belongs to a true electrochemical response. For consecutive suspected true mechanism frequencies, the data processing server forms protection verification segments according to the sequential position of the frequencies. These protection verification segments are fragment results composed of consecutive suspected true mechanism frequency indexes, used for reading and verification by frequency band in step S4.

[0092] The data processing server then generates candidate processing spectra based on the point-level spectral record sequence. The candidate processing spectra are constructed based on the original impedance spectrum. Frequency points in the temporary rejection state are temporarily excluded from the full-spectrum consistency calculation in the candidate processing spectrum's participation list. Frequency points in the reduced-weight state are retained in the participation list with their reduced-weight coefficients. Frequency points in the marked state retain their original data and carry anomaly source information. Frequency points in the protection and verification state are retained in the participation list and carry protection and verification segment information. Frequency points in the retained frequency point set are entered into the candidate processing spectrum according to their original frequency, real part of impedance, imaginary part of impedance, and phase. After the candidate processing spectrum is generated, the data processing server synchronously outputs a processing action index table, trial rejection segments, and protection and verification segments, enabling step S4 to read the candidate processing spectrum by frequency point index, cancel trial rejection, or retain the reduced-weight and marked results.

[0093] After processing in step S3, the processing branch output in step S2 is converted into executable frequency point processing actions. Strong candidate abnormal frequency points are temporarily removed, suspected disturbance frequency points are either downweighted or marked, suspected true mechanism frequency points are protected and verified, and retained frequency points are added to the candidate processing spectrum. Step S3 simultaneously outputs the candidate processing spectrum, processing action index table, trial removal segments, and protection and verification segments, providing direct calling objects for step S4 to conduct full-spectrum quality verification, cancel trial removal, and generate abnormal processing records.

[0094] Step S4: Perform full-spectrum quality verification based on candidate processed spectra, and confirm the corresponding processing actions based on changes in full-spectrum consistency and key electrochemical parameters; cancel trial rejection if no confirmation is made, retain the weighting and labeling results, and output the processed impedance spectrum and abnormal processing records.

[0095] In this embodiment, step S4 uses the candidate processing spectrum, processing action index table, trial rejection segment, protection verification segment formed in step S3, and the spectral deviation, sampling stability deviation, and processing branch recorded in step S2 as input. Trial rejection, weight reduction, marking, and protection verification in the candidate processing spectrum are still processing results to be confirmed. Step S4 performs a full-spectrum quality verification on the candidate processing spectrum, confirms the processing action or cancels the trial rejection based on the changes in full-spectrum consistency and key electrochemical parameters, and simultaneously generates the processed impedance spectrum and abnormal processing record.

[0096] S401: Construct the full-spectrum quality verification input spectrum.

[0097] The data processing server initiates processing from the candidate processed spectrum output in step S3, and simultaneously reads the processing action index table, trial rejection segments, protection verification segments, and point-level spectral record sequence. After reading, the data processing server establishes a correspondence between the original impedance spectrum and the candidate processed spectrum using the frequency index as the unique reference key, enabling full-spectrum quality verification to compare the changes in spectral data before and after processing at the same frequency source.

[0098] The data processing server first constructs an original verification spectrum based on the point-level spectral record sequence. The original verification spectrum is a verification object formed by all frequency points obtained in step S1 and written into the evidence results in step S2. Each frequency point carries its frequency, real part of impedance, imaginary part of impedance, and phase according to its frequency point order, and retains the corresponding spectral deviation, sampling stability deviation, and processing branch. Subsequently, the data processing server constructs a candidate verification spectrum based on the candidate processing spectrum. In the candidate verification spectrum, frequency points in the temporarily removed state are not included in the full-spectrum consistency calculation; frequency points in the reduced-weight state are retained and their calculation contribution is reduced according to the reduction coefficient in the processing action index table; frequency points in the marked state retain their original data and carry anomaly source prompts; and frequency points in the protected verification state are retained and carry protected verification segment information.

[0099] In the impedance spectroscopy detection of corrosion samples, if step S3 writes one mid-frequency point to the temporary rejection state and two low-frequency points to the weighted state, the data processing server generates two verification inputs in S401: the original verification spectrum still contains all frequency points, and the candidate verification spectrum does not use the mid-frequency temporarily rejected frequency point in the full spectrum consistency calculation, while retaining the two low-frequency weighted frequency points. Both verification inputs are backlinked to the same set of point-level spectral point records through frequency point indexes, and the verification results obtained from subsequent comparisons can be written back to the corresponding frequency points.

[0100] S402: Calculate the variation in full spectrum consistency.

[0101] After the original and candidate verification spectra are formed, the data processing server performs full-spectrum consistency calculations on both verification inputs. The full-spectrum consistency change represents the degree of improvement of the candidate processed spectrum relative to the original impedance spectrum in the overall trend of the frequency sequence. In this implementation, the trend consistency residual between adjacent frequency bands is used as the default verification amount for the full-spectrum consistency change. When the electrochemical workstation or data processing server is configured with a Kramers-Kronig consistency calculation module or an equivalent circuit fitting model, the trend consistency residual between adjacent frequency bands can be replaced with the Kramers-Kronig consistency residual or the equivalent circuit fitting residual under the same frequency participation rule, but only one residual calculation method is used for the same full-spectrum quality verification. When using the Kramers-Kronig consistency residual, the data processing server performs the same Kramers-Kronig consistency test on both the original and candidate verification spectra. Specifically, the data processing server first determines the set of participating frequencies according to the frequency participation rules, temporarily removing state frequencies from the set, and retaining the weighting coefficients of downweighted state frequencies. Then, the participating frequencies are arranged in frequency order and converted to angular frequencies. Subsequently, the data processing server sets multiple relaxation time constants at logarithmic intervals within the frequency range covered by the participating frequencies, and jointly fits the real and imaginary parts of the impedance of the participating frequencies using a linear Kramers-Kronig consistency model composed of series resistance terms and multiple parallel relaxation terms. The original and candidate complex kernel spectra use the same number of relaxation time constants, the same regularization parameters, the same initial value range, and the same convergence conditions. After fitting, the data processing server obtains the predicted real and imaginary impedance values ​​for each participating frequency, and uses the sum of the absolute values ​​of the real and imaginary impedance prediction errors as the Kramers-Kronig consistency error term for that frequency. When fitting the residuals using an equivalent circuit, if the impedance spectroscopy test task has a pre-configured equivalent circuit model, the data processing server reads this pre-configured equivalent circuit model and fits the original kernel spectrum and candidate kernel spectrum using the same model, the same initial parameter range, and the same convergence conditions. If the impedance spectroscopy test task has not pre-configured an equivalent circuit model but still selects to fit the residuals using an equivalent circuit, the data processing server defaults to using an equivalent circuit model with a solution resistor connected in series with a charge transfer branch, where the charge transfer branch is composed of a charge transfer resistor connected in parallel with a constant-phase element. When low-frequency candidate frequencies exhibit diffusion tailing characteristics, a diffusion impedance element is further connected in series in the default model. After fitting, the data processing server uses the magnitude deviation between the fitted impedance and the measured impedance at the participating frequencies as the fitting error term for that frequency.The Kramers-Kronig consistency error terms or equivalent circuit fitting error terms mentioned above are all summarized according to the contribution ratio of the participating frequency points. Temporarily excluded state frequency points are not included in the summary. Reduced weight state frequency points are included in the summary according to the reduction coefficient Wi. Retained frequency points, marked state frequency points, and protection verification state frequency points are included in the summary according to the contribution ratio 1.

[0102] When using the adjacent frequency band trend consistency residual, the data processing server sequentially reads the real part of the impedance, the imaginary part of the impedance, and the logarithmic frequency position of each participating frequency point, calculates the normalized difference vector of adjacent frequency bands, and uses the directional change between two adjacent normalized difference vectors as the local trend residual for that frequency point. For the i-th participating frequency point, the local trend residual... Calculate using the following formula: ; in, Let be the local trend residual of the i-th participating frequency point, used to represent the change in adjacent trends of that frequency point when participating in the full spectrum consistency calculation; and These are the normalized difference vectors on the left and right sides of the i-th frequency point, respectively, derived from the frequency point sequence involved in the full-spectrum consistency calculation. The full-spectrum consistency residual of the original verification spectrum. The contribution ratio of each participating frequency point is summarized, and the contribution ratio of all participating frequencies in the original verification spectrum is set to 1; the full spectrum consistency residual of the candidate verification spectrum is obtained by summing the contribution ratios of each participating frequency point. The results are summarized based on the contribution ratio of the participating frequency points in the candidate review spectrum. Temporarily excluded state frequency points are not included in the summary. Reduced weighted state frequency points are calculated according to the weighting coefficient. In summary, other participating frequencies are summarized according to their contribution ratio: ; in, For candidate full-spectrum consistency residuals; This is the set of frequency point indices in the candidate verification spectrum that participate in the full spectrum consistency calculation; The contribution ratio of the i-th frequency point is determined by the frequency point in the deweighted state. The frequency points for retention, marking status, and protection verification status are all set to 1. Let be the local trend residual at the i-th frequency point. Original full-spectrum consistency residual. Calculated using the same formula, and all participating frequencies in the original complex spectrum. Take 1.

[0103] Improvement in overall spectrum consistency We obtain the following formula: ; in, The original full-spectrum consistency residual is derived from the adjacent frequency band trend consistency residual, Kramers-Kronig consistency residual, or equivalent circuit fitting residual of the original composite spectrum. The candidate full-spectrum consistency residual is derived from the residuals obtained by the candidate verification spectra under the same calculation method. This represents the improvement in overall spectrum consistency, indicating the percentage decrease in the residual of the candidate processed spectrum relative to the original impedance spectrum. When... When the value is 0, the data processing server does not calculate the improvement in full-spectrum consistency and generates a result indicating no improvement in full-spectrum consistency; when... When greater than 0, press Calculate the improvement in overall spectrum consistency.

[0104] The consistency improvement discrimination boundary is formed by recalculating the single-point leave-one-out residual of the retained frequency point set. The data processing server temporarily removes one retained frequency point from the original verification spectrum in turn, and obtains the leave-one-out residual in the same way as the consistency residual of the original full spectrum, and calculates the change ratio of the leave-one-out residual relative to the consistency residual of the original full spectrum; the change ratios corresponding to all retained frequency points form the residual fluctuation sequence, and the consistency improvement discrimination boundary is taken as the upper quartile of the sequence in ascending order. When the result exceeds the consistency improvement threshold, a full-spectrum consistency improvement result is generated. If the result is not greater than the consistency improvement threshold, a result indicating no improvement in full-spectrum consistency is generated.

[0105] As an example, in the impedance spectrum detection of fuel cells, after temporarily removing a certain isolated abrupt change frequency point, the trend consistency residual of adjacent frequency bands of the candidate verification spectrum decreases. The data processing server obtains the full spectrum consistency improvement amount from the original full spectrum consistency residual and the candidate full spectrum consistency residual, and writes the comparison result into the processing action index table.

[0106] S403: Determine the rationality of changes in key electrochemical parameters.

[0107] After the uniform changes across the entire spectrum are established, the data processing server calculates key electrochemical parameters based on the original composite spectrum and candidate composite spectra, respectively. Key electrochemical parameters are a set of parameters used to characterize the main response features of the tested electrochemical system. In this embodiment, key electrochemical parameters include at least solution resistance and charge transfer resistance; when the impedance spectroscopy test pre-specifies an equivalent circuit model, the data processing server also reads capacitance-related parameters or diffusion-related parameters from the fitting results of the same equivalent circuit model and incorporates them into the determination of the reasonableness of parameter changes.

[0108] Without a pre-specified equivalent circuit model, the data processing server extracts at least the solution resistance and charge transfer resistance as key electrochemical parameters. The stability intercept here refers to the estimated intercept along the real axis of the complex impedance plane after the real part of the high-frequency impedance enters a relatively stable plateau, not the artificially extrapolated intercept. The data processing server reads the top 10% of participating frequencies in descending order as candidate frequencies for the high-frequency end; when the number of participating frequencies is less than 30, it reads the three highest-frequency participating frequencies as candidate frequencies for the high-frequency end; subsequently, it selects a group of consecutive frequencies with the smallest adjacent difference in the real part of the impedance from the candidate frequencies for the high-frequency end, and takes the median level of the real part of the impedance of this group of consecutive frequencies as the solution resistance. The mid-to-low frequency arc here refers to the consecutive candidate frequency bands located in the bottom 60% range after being sorted from high to low frequency, which are not selected as candidate frequencies for the high-frequency end, and are used to characterize the span of the real part of the impedance of the charge transfer arc or its low-frequency extension. The data processing server reads the maximum real part of the impedance from the candidate frequency points in the mid-to-low frequency band as the upper edge of the real part of the impedance, and uses the difference between the upper edge of the real part of the impedance and the solution resistance as the charge transfer resistance. If an equivalent circuit model has been pre-configured for this test, the data processing server fits the original complex spectrum and the candidate complex spectrum with the same equivalent circuit model, and reads the solution resistance, charge transfer resistance, capacitance-related parameters, or diffusion-related parameters from the fitting results.

[0109] The rationality of parameter changes is determined by the proportion of the parameter change. For the j-th key electrochemical parameter, the data processing server calculates the proportion of parameter change. : ; in, This represents the percentage change in the j-th key electrochemical parameter. This is the j-th key electrochemical parameter obtained from the original composite spectrum; This is the j-th key electrochemical parameter obtained from the candidate composite spectrum; To prevent A tiny positive number that is 0, which makes division impossible; its value is one-thousandth of the median effective value of similar key electrochemical parameters. Used to determine whether candidate treatment spectra cause key electrochemical parameters to shift beyond reasonable boundaries.

[0110] The reasonable boundary for parameter variation is formed by recalculating the parameters at a single point in the retained frequency set. The data processing server sequentially removes one retained frequency point from the original composite spectrum and recalculates the key electrochemical parameters according to the same extraction rules as the original key electrochemical parameters. The percentage change of the key electrochemical parameter obtained in each recalculation relative to the original key electrochemical parameter is used as the parameter perturbation record corresponding to that retained frequency point. The reasonable boundary for parameter variation is taken as the 90th percentile of the parameter perturbation records in ascending order. All key electrochemical parameters... When none of the parameters exceed the reasonable variation boundary of the corresponding parameter, reasonable parameter variation results are generated; for any key electrochemical parameter When the change exceeds the reasonable boundary of the corresponding parameter, an abnormal result of parameter change is generated.

[0111] For the protected verification segment, the data processing server reads the spectral deviation, physical evidence status, and frequency sequence position of the frequency points in the protected verification segment separately, and compares the fitting response of the frequency band where the protected verification segment is located in the original verification spectrum and the candidate verification spectrum. If the change of key electrochemical parameters after the protected verification segment is retained is within the reasonable parameter change boundary, the data processing server confirms the retention of the protected verification segment; if the trial removal of the segment causes the key electrochemical parameters to exceed the reasonable parameter change boundary, even if the full spectrum consistency residual of the candidate verification spectrum decreases, the data processing server still lists the trial removal segment as an object to be withdrawn. As an example, in battery impedance spectroscopy detection, a certain low-frequency continuous bending frequency band is determined by step S2 to be a suspected true mechanism. In S403, the data processing server retains this frequency band to participate in the equivalent circuit fitting. If the candidate key electrochemical parameters do not exceed the reasonable parameter change boundary, the protected verification segment is confirmed to be retained.

[0112] S404: Confirm the processing action and output the processed impedance spectrum and anomaly handling record.

[0113] After obtaining the improved overall spectrum consistency and reasonable parameter changes, the data processing server confirms or adjusts the processing actions formed in step S3 based on the two types of verification results. The confirmation conditions are: the candidate processed spectrum generates an improved overall spectrum consistency result relative to the original impedance spectrum, and the key electrochemical parameters generate reasonable parameter changes. When the confirmation conditions are met, the data processing server confirms the trial removal, weight reduction, marking, and protection verification results in the processing action index table. Frequency points in the temporary removal state are not included in the processed impedance spectrum; frequency points in the weight reduction and marking states are retained in the processed impedance spectrum; and frequency points in the protection verification state are retained according to the original impedance spectrum data.

[0114] When the overall spectrum consistency does not improve or abnormal parameter changes occur, the data processing server reads the trial rejection segment and restores the corresponding temporary rejection status to a retained participation relationship according to the frequency point index, generating a revocation trial rejection result. The revocation trial rejection result indicates that the trial rejection action was not confirmed after the overall spectrum quality review. It does not change the deweighting status, marking status, or protection review status; the deweighting status and marking status remain in the processed impedance spectrum, and the frequency points corresponding to the protection review status continue to be retained according to the original data.

[0115] When multiple trial rejection segments exist in the candidate processing spectrum, the data processing server performs a review on each trial rejection segment one by one. For any trial rejection segment, the data processing server first maintains the temporary rejection state of the trial rejection segment and calculates the corresponding full spectrum consistency result and parameter change result. Then, it restores the participation relationship of the trial rejection segment and recalculates the full spectrum consistency result and parameter change result. If the abnormal parameter change result is eliminated after restoring the trial rejection segment, or if the full spectrum consistency improvement result cannot be generated when maintaining the trial rejection segment, then the trial rejection segment generates a cancellation result. If the full spectrum consistency improvement result can be generated when maintaining the trial rejection segment and the parameter change result is reasonable, and the full spectrum consistency improvement amount decreases after restoring the trial rejection segment, then the trial rejection segment maintains the temporary rejection state and confirms the trial rejection action. After the restoration is completed, the data processing server regenerates the processed impedance spectrum. The processed impedance spectrum consists of the confirmed retained frequency points, the reduced weight state frequency points, the marked state frequency points, the protection review state frequency points, and the frequency points restored after the cancellation of the trial rejection, and outputs them in the order of the frequency points formed in step S1.

[0116] The data processing server simultaneously generates anomaly handling records. These records use the frequency index as the primary key, reading spectral deviation, sampling stability deviation, and processing branches from the point-level spectral record; reading processing actions from the processing action index table; reading the full-spectrum consistency improvement result from S402; reading the reasonable parameter change result from S403; and reading the result of whether the trial rejection was revoked from S404. All of the above information is then written into the record item corresponding to the same frequency index. As an example, in coating impedance spectroscopy detection, if a strong candidate abnormal frequency point improves full-spectrum consistency and reasonable changes in key electrochemical parameters after trial rejection, the anomaly handling record states that the processing action for that frequency point was trial rejection and that the trial rejection was not revoked. If another strong candidate abnormal frequency point causes an abnormal shift in charge transfer resistance after trial rejection, the data processing server restores the participation relationship of that frequency point and states in the anomaly handling record that the trial rejection has been revoked.

[0117] After step S4, the candidate processed spectra undergo full-spectrum quality verification. Trial elimination actions are confirmed or revoked based on changes in full-spectrum consistency and key electrochemical parameters. Deweighting, labeling, and protection verification results are retained along with the frequency index. Step S4 ultimately outputs the processed impedance spectrum and anomaly handling records. The processed impedance spectrum is used for subsequent equivalent circuit fitting, full-spectrum consistency verification, or electrochemical diagnosis. The anomaly handling records retain the evidence source, processing action, verification results, and trial elimination revocation status for each frequency point.

[0118] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for automatically eliminating impedance spectral anomalies in an electrochemical workstation, characterized in that, include: Step S1: Obtain impedance spectral point data and acquisition quality information generated during the impedance spectral test at the electrochemical workstation, and associate the impedance spectral point data and acquisition quality information into point-level spectral point records according to the same frequency point index; Step S2: Based on point-level spectral point records, generate spectral curve deviation to characterize local spectral shape changes and acquisition stability deviation to characterize acquisition stability changes. Combine the acquisition evidence state and physical evidence state, and output the processing branch for each frequency point through the sequence state discrimination model. Step S3: Based on the processing branch, perform the corresponding processing actions in trial elimination, weight reduction, marking and protection review on the corresponding frequency points, and synchronize the processing actions with the frequency point index to generate candidate processing spectrum; Step S4: Perform full-spectrum quality verification based on candidate processed spectra, and confirm the corresponding processing actions based on changes in full-spectrum consistency and key electrochemical parameters; cancel trial rejection if no confirmation is made, retain the weighting and labeling results, and output the processed impedance spectrum and abnormal processing records.

2. The method for automatically eliminating impedance spectrum anomalies in an electrochemical workstation according to claim 1, characterized in that, In step S1, the data processing server generates frequency indexes with the same test batch as the reading boundary, and connects the impedance spectrum data and acquisition quality information according to the frequency indexes; frequency points that do not match acquisition quality information are written as acquisition quality missing markers, and the matched records form a point-level spectrum record sequence according to the logarithm of the frequency.

3. The method for automatically eliminating impedance spectrum anomalies in an electrochemical workstation according to claim 2, characterized in that, In step S2, the data processing server forms the normalized trend of adjacent frequency bands based on the real part of impedance, the imaginary part of impedance, and the logarithmic frequency position of adjacent frequency points in the point-level spectral point recording sequence, and obtains the local turning angle from the directional change between adjacent normalized trends. The data processing server uses the median level and median deviation of the local corners within the adjacent frequency window as local geometric references to normalize the degree of local corner deviation at the current frequency point, obtain the spectral deviation, and write the physical evidence status based on the comparison result between the spectral deviation and the spectral deviation discrimination boundary.

4. The method for automatically eliminating impedance spectrum anomalies in an electrochemical workstation according to claim 3, characterized in that, In step S2, the sampling stability deviation is determined by response stability, phase stability, repeatability deviation, and impedance magnitude. The data processing server obtains the normalized repeatability deviation based on the impedance magnitude, extracts the stable acquisition reference from the stable range frequency point, and writes the sampling stability deviation into the acquisition evidence status after comparing it with the stable acquisition reference.

5. The method for automatically eliminating impedance spectrum anomalies in an electrochemical workstation according to claim 4, characterized in that, In step S2, the sequence state discrimination model is implemented using a hidden Markov model. The hidden Markov model forms the observation sequence with spectral bias and sampling stability bias, and uses the retained, strong candidate anomalies, suspected sampling disturbances and suspected real mechanisms as hidden states, and processes the processing branch through the maximum probability hidden state path output.

6. The method for automatically eliminating impedance spectrum anomalies in an electrochemical workstation according to claim 5, characterized in that, In step S3, the data processing server generates a processing action index table based on the processing branches; the processing branches form a temporary removal state for frequency points with strong candidate anomalies, the processing branches form a corresponding state in the deweighting state and the marking state for frequency points suspected of having disturbances, the processing branches form a protection and verification state for frequency points suspected of having the true mechanism, and the processing branches enter the candidate processing spectrum according to the original data for the retained frequency points.

7. The method for automatically eliminating impedance spectrum anomalies in an electrochemical workstation according to claim 6, characterized in that, In step S3, the weighting state determines the weighting coefficient based on the degree of deviation of the sampling stability deviation from the sampling stability deviation discrimination boundary, and binds the weighting coefficient to the corresponding frequency point index; The marker status is generated when the range status is in the range change neighborhood, the evidence collection status is insufficient evidence collection, or there is a marker indicating missing collection quality, and it carries an anomaly source prompt.

8. The method for automatically eliminating impedance spectrum anomalies in an electrochemical workstation according to claim 7, characterized in that, In step S4, the data processing server constructs the original verification spectrum based on the point-level spectral point record sequence, and constructs the candidate verification spectrum based on the candidate processing spectrum; In the candidate verification spectrum, frequency points in the temporary elimination state are not included in the full spectrum consistency calculation, frequency points in the reduced weight state are included in the full spectrum consistency calculation according to the reduction weight coefficient, and frequency points in the marked state and the protected verification state retain the original data.

9. The method for automatically eliminating impedance spectrum anomalies in an electrochemical workstation according to claim 8, characterized in that, In step S4, the full spectrum quality verification includes the calculation of full spectrum consistency changes; the data processing server uses the same residual calculation method to obtain the original full spectrum consistency residual and the candidate full spectrum consistency residual for the original verification spectrum and the candidate full spectrum consistency residual, and forms the full spectrum consistency improvement result based on the original full spectrum consistency residual and the candidate full spectrum consistency residual, and reads the contribution ratio of each participating frequency point using the residual calculation method.

10. The method for automatically eliminating impedance spectrum anomalies in an electrochemical workstation according to claim 9, characterized in that, In step S4, the full-spectrum quality verification also includes the determination of changes in key electrochemical parameters; the data processing server extracts key electrochemical parameters from the original verification spectrum and the candidate verification spectrum respectively, forms reasonable results or abnormal results of parameter changes, and generates anomaly processing records based on the full-spectrum consistency improvement results and parameter change determination results.

Images