A multi-agent based autonomous analysis method for spectral data

Abstract

The application provides a kind of multi-agent-based spectral data autonomous analysis method, it is related to the technical field of data processing, method includes: through multi-agent cooperation mechanism, the whole process of spectral data from acquisition, quality evaluation, pretreatment, spectral peak analysis to phase inference is unified modeling and scheduling, data quality information is converted into constraintable pretreatment strategy by central coordination agent, and consistency check and conflict arbitration are implemented on spectral peak parameters and phase inference results at the end of analysis, so that adaptive processing and result check are realized under the premise of maintaining spectral structure correlation relationship.The application can improve the accuracy of spectral data analysis.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a method for autonomous analysis of spectral data based on multi-agent systems. Background Technology

[0002] Existing spectral data analysis techniques typically employ a processing model centered on human experience or a single algorithmic workflow. In practical applications, data acquisition, preprocessing, peak extraction, and phase identification are generally performed sequentially in a fixed order. These processes often rely on manually set parameters or the independent operation of rule-based algorithm modules. In such solutions, spectral data quality assessment is often only used as an auxiliary judgment before preprocessing, failing to provide continuous constraints on subsequent analysis procedures. Preprocessing methods typically employ uniform algorithm combinations and fixed intensity parameters, lacking the ability to adaptively adjust to differences in spectral quality. Peak analysis and phase identification processes are mostly independent of each other, lacking cross-stage information feedback and result verification mechanisms. Consequently, in complex samples, low signal-to-noise ratio data, or multi-phase scenarios, unstable analytical results, insufficient interpretability, and reliance on repeated manual corrections are prone to occur.

[0003] The lack of a unified and comprehensive mechanism in existing technologies to coordinate and constrain spectral data quality, preprocessing strategies, peak analysis results, and phase inference results leads to a fragmented analytical workflow. Specifically, this manifests as: the preprocessing stage potentially disrupting key peak structures; peak analysis results failing to provide a consistent interpretation with phase determination; a lack of systematic arbitration for conflicts between different phase candidates; and difficulty in tracing the formation path and source of evidence for the final analytical conclusions. These deficiencies are particularly pronounced in applications involving multi-source spectral data, high-noise environments, or competition among multiple candidate phases, making it difficult to meet the practical requirements of simultaneously demanding analytical accuracy. Summary of the Invention

[0004] This invention provides a multi-agent-based autonomous analysis method for spectral data, which can improve the accuracy of spectral data analysis.

[0005] A first aspect of the present invention provides a method for autonomous analysis of spectral data based on multiple agents, the method comprising: Obtain the raw spectral data corresponding to the target analysis object and bind it with a preset identifier to form a raw spectral data set; The quality diagnostic intelligent agent performs data quality assessment processing on the original spectral data set to generate data quality assessment results corresponding to each original spectral data. The central coordinating agent decomposes the analysis task based on the data quality assessment results and determines the preprocessing constraints and processing intensity boundaries that match the data quality assessment results, thereby generating a set of preprocessing strategies corresponding to the quality-labeled spectral data set. The quality-labeled spectral data set is formed by binding the data quality assessment results with the original spectral data set. Based on the set of preprocessing strategies, spectral preprocessing is performed on the original spectral data set to obtain a preprocessed spectral data set, wherein the structural relationship between the preprocessed spectral data set and the original spectral data set is maintained during the preprocessing process; Based on the preprocessed spectral data set, the peak fitting agent performs peak analysis processing on each preprocessed spectral data to generate a set of peak parameters corresponding to each preprocessed spectral data. The phase identification agent matches the set of spectral peak parameters with the template information in the multi-source reference spectral library, and combines the sample prior information with the confidence weight of the spectral library to generate a set of phase inference results corresponding to each candidate. The central coordinating agent performs consistency verification and conflict arbitration on the set of spectral peak parameters and the set of phase inference results to form the final set of analysis results, which is then output as structured analysis results.

[0006] Based on the above technical solutions, preferably, the step of performing data quality assessment processing on the original spectral data set by the quality diagnostic intelligent agent to generate data quality assessment results corresponding to each original spectral data specifically includes: A quality assessment context is established for each of the original spectral data, and the quality assessment context includes at least an acquisition condition identifier, an instrument status identifier, and a time identifier; Under the constraints of the quality assessment context, the original spectral data are jointly modeled based on the global amplitude distribution, local fluctuation distribution, and noise ratio of the spectral intensity sequence to extract signal-to-noise features that characterize the level of random noise. By fitting the slowly varying trend term within a preset spectral axis range to each original spectral data, and evaluating the continuity and consistency of the fitting residual as the spectral axis position changes, a baseline stability feature is generated to characterize the degree of baseline drift. By analyzing the intensity abrupt change relationship between sampling points on adjacent spectral axes and the probability of repeated occurrence of abrupt change points within a local window, spike noise features used to characterize isolated spikes and outlier perturbations are extracted from the original spectral data. The original spectral data are processed by detecting whether the spectral intensity reaches the upper limit of the instrument range within the continuous spectral axis range, and spectral saturation features are generated to characterize the risk of effective information truncation. Based on the alignment relationship between multiple original spectral data in the original spectral data set, the relative position deviation of characteristic peaks in different original spectral data is analyzed to generate spectral axis consistency features to characterize the stability of the spectral axis. The signal-to-noise ratio feature, baseline stability feature, spike noise feature, spectral line saturation feature, and spectral axis consistency feature are mapped to a unified evaluation scale and assigned corresponding weight parameters according to a preset quality evaluation strategy to generate data quality evaluation results corresponding to each of the original spectral data.

[0007] Based on the above technical solutions, preferably, the central coordinating agent decomposes the analysis task according to the data quality assessment results and determines the preprocessing constraints and processing intensity boundaries that match the data quality assessment results, thereby generating a set of preprocessing strategies corresponding to the quality-labeled spectral data set, specifically including: The data quality assessment results are parsed to identify the corresponding quality feature type, anomaly degree, and repair level in each data quality assessment result, thereby generating a corresponding quality constraint description for each of the original spectral data. Based on the quality constraint description, the overall spectral analysis task is decomposed into preprocessing subtasks associated with data quality characteristics. Each of the preprocessing subtasks is bound to a corresponding quality trigger condition, so that each of the preprocessing subtasks is activated when the corresponding quality anomaly condition is met; Based on the quality constraint description, preprocessing constraints are determined for each preprocessing subtask. These preprocessing constraints are used to define the boundaries by which the preprocessing subtask affects the spectral structure during execution. Based on the quantification level of each quality feature in the data quality assessment results, a processing intensity boundary is determined for each preprocessing subtask. The processing intensity boundary is used to limit the range of parameter adjustment in the corresponding preprocessing subtask. The preprocessing subtasks, preprocessing constraints, and processing intensity boundaries are jointly scheduled and arranged according to the data dependencies and structural influence order among the preprocessing subtasks, and a preprocessing decision scheme is generated. Based on the preprocessing decision scheme, the preprocessing subtasks, the preprocessing constraints, and the processing intensity boundaries are combined to generate a set of preprocessing strategies corresponding to the quality-labeled spectral data set.

[0008] Based on the above technical solutions, preferably, the step of performing spectral preprocessing on the original spectral data set according to the preprocessing strategy set to obtain a preprocessed spectral data set specifically includes: A preprocessing execution context is established based on the preprocessing strategy set, and the preprocessing execution context is used to limit the execution order of preprocessing subtasks; Under the constraints of the preprocessing execution context, spectral preprocessing operations are performed sequentially for each of the original spectral data in the execution order. When performing the baseline correction preprocessing subtask, the slowly varying trend term is modeled and corrected according to the preprocessing constraints. When performing the noise suppression preprocessing subtask, random noise components are suppressed in a restricted manner according to the processing intensity boundary; When performing the peak repair preprocessing subtask, local replacement or reconstruction processing is performed on abnormal sampling points; When performing the spectral axis alignment adjustment preprocessing subtask, normalization is performed without changing the number and arrangement order of spectral axis sampling points; After completing each preprocessing subtask, the processed spectral data are summarized in a predetermined order to generate the preprocessed spectral data set.

[0009] Based on the above technical solutions, preferably, the step of using a peak fitting agent to perform peak analysis processing on each preprocessed spectral data based on the preprocessed spectral data set to generate a set of spectral peak parameters corresponding to each preprocessed spectral data set specifically includes: To establish a spectral peak analysis context based on the preprocessed spectral data set, the spectral peak analysis context shall at least include spectral axis index mapping relationships, key structural position mapping relationships, and preprocessing process records; Under the constraints of the spectral peak analysis context, initial spectral peak detection processing is performed on each preprocessed spectral data. By jointly analyzing the local extreme value features, gradient change features, and curvature change features of the spectral intensity along the spectral axis, a set of potential spectral peak candidate positions is generated. Based on the spectral morphology information in the peak analysis context, an adaptive peak number determination process is performed on each preprocessed spectral data. By analyzing the spacing relationship, overlap degree, and local signal-to-noise level between the candidate peak positions in the candidate peak position set, the target peak number corresponding to each preprocessed spectral data is determined. Initial spectral peak parameters are generated for each target spectral peak. These initial spectral peak parameters include peak position parameters, peak width parameters, peak shape parameters, and peak intensity parameters. The value range of the initial spectral peak parameters is jointly limited by the structure preservation constraints in the spectral peak analysis context and the preprocessing process records. The initial spectral parameters are iteratively adjusted by constructing an objective function that includes a spectral peak superposition model and a background term for each of the preprocessed spectral data, and a robust loss constraint is introduced during the fitting process to output the fitting result. The fitting results are compared with the preprocessed spectral data in terms of overall contour, local residual distribution and key structural positions. When the spectral peak analysis results meet the preset reliability conditions, the final spectral peak parameters are output. A set of spectral peak parameters corresponding to each of the preprocessed spectral data is generated for the final spectral peak parameters.

[0010] Based on the above technical solutions, preferably, the step of the phase identification agent matching the set of spectral peak parameters with the template information in the multi-source reference spectral library, and combining the sample prior information and the confidence weight of the spectral library to generate a set of phase inference results corresponding to each candidate, specifically includes: The multi-source reference spectrum library is used to select a set of candidate templates that are comparable to the set of spectral peak parameters in terms of structural features based on the distribution range of peak position parameters, peak number characteristics and spectral band coverage relationship. By establishing the correspondence between the peak position parameters in the spectral peak parameter set and the characteristic peak positions of the candidate templates in the candidate template set, a peak position alignment mapping relationship is formed; Based on the peak alignment mapping relationship, the matching measurement results are generated for each candidate template by jointly calculating the peak position deviation, peak width matching degree, peak shape similarity, and peak intensity ratio consistency. The matching metric results are constrained and corrected by introducing prior information of the sample. A first template is obtained by suppressing or eliminating candidate templates that do not conform to the prior information of the sample. The first template is weighted and fused by introducing spectral library confidence weights to obtain the second template; A comprehensive inference process is performed on the candidate phases corresponding to the second template to generate phase inference results corresponding to each candidate phase, forming the set of phase inference results.

[0011] Based on the above technical solutions, preferably, the step of the central coordinating agent performing consistency verification and conflict arbitration processing on the set of spectral peak parameters and the set of phase inference results to form a final set of analysis results, and outputting it as a structured analysis result, specifically includes: Construct corresponding arbitration contexts for each of the candidate phases; Under the arbitration context constraints, the spectral peak parameter set is subjected to spectral peak evidence reconstruction processing and mapped to a spectral peak evidence set; A consistency verification process is performed on the set of phase inference results to generate a consistency verification record corresponding to each candidate. The consistency verification process includes peak consistency verification for key peak sets, prior consistency verification for sample prior information, and library consistency verification for the stability of multi-source reference spectral libraries. Based on the consistency verification record, the conflict relationship between candidate phases is explicitly modeled. By identifying the competitive relationship between candidate phases around the same key spectral peak set, the mutual exclusion relationship between different key spectral peak sets, and the incompatibility relationship between candidate phases and sample prior information, a conflict graph structure is constructed. Based on the conflict graph structure, conflict arbitration processing is performed to form an arbitration ranking result of candidate phases. The conflict arbitration processing ranks the candidate phases in the order of prior decision, spectral peak decision, and spectral library decision. Based on the arbitration ranking results, a backtracking verification process is performed on the first preset number of candidate phases. This involves remapping the corresponding candidate templates to the spectral axis index mapping relationship of the preprocessed spectral data set and verifying whether the interpretation residuals at the key spectral peak sets meet the preset confidence conditions. After the backtracking verification is passed, the phase inference results that have passed the arbitration are encapsulated in a structured manner to form the final set of analysis results.

[0012] In a second aspect of the invention, a multi-agent-based autonomous spectral data analysis device is provided. The device is used to execute a multi-agent-based autonomous spectral data analysis method as described in any of the preceding embodiments. The device includes an acquisition module, a processing module, and an output module, wherein: The acquisition module is used to acquire the original spectral data corresponding to the target analysis object and bind a preset identifier to form a set of original spectral data. The processing module is used to control the quality diagnostic agent to perform data quality assessment processing on the original spectral data set and generate data quality assessment results corresponding to each original spectral data. The processing module is used to control the central coordinating agent to decompose the analysis task according to the data quality assessment result, and determine the preprocessing constraints and processing intensity boundaries that match the data quality assessment result, thereby generating a set of preprocessing strategies corresponding to the quality-labeled spectral data set, wherein the quality-labeled spectral data set is formed by binding the data quality assessment result with the original spectral data set; The processing module is used to perform spectral preprocessing on the original spectral data set according to the preprocessing strategy set to obtain a preprocessed spectral data set, wherein the structural correlation between the preprocessed spectral data set and the original spectral data set is maintained during the preprocessing process; The processing module is used to control the peak fitting agent to perform peak analysis processing on each preprocessed spectral data based on the preprocessed spectral data set, and generate a set of spectral peak parameters corresponding to each preprocessed spectral data. The processing module is used to control the phase identification agent to match the set of spectral peak parameters with the template information in the multi-source reference spectral library, and combine the sample prior information and the confidence weight of the spectral library to generate a set of phase inference results corresponding to each candidate. The output module is used to control the central coordinating agent to perform consistency verification and conflict arbitration on the set of spectral peak parameters and the set of phase inference results, form the final set of analysis results, and output them as structured analysis results.

[0013] In a third aspect of the invention, an electronic device is provided, including a processor, a memory, a user interface, and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any of the preceding embodiments.

[0014] In a fourth aspect of the invention, a non-transitory computer-readable storage medium is provided, the computer-readable storage medium storing instructions that, when executed, perform the method as described in any of the preceding claims.

[0015] In summary, one or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages: This invention introduces a quality diagnosis agent, a peak fitting agent, a phase identification agent, and a central coordination agent to collaboratively model and unify constraints the entire process of spectral data analysis, from quality assessment, preprocessing, peak analysis to phase inference. This ensures that data quality information continues to influence subsequent analysis stages. Furthermore, consistency checks and conflict arbitration mechanisms cross-validate peak parameters and phase inference results, preventing key peaks from being destroyed during preprocessing, reducing interference from noise and anomalies in peak analysis, and suppressing misjudgments caused by inconsistent evidence among multiple candidate phases. Overall, this invention achieves adaptive adjustment and result verification of the analysis process, significantly improving the accuracy, stability, and reliability of spectral data analysis results in complex scenarios. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a multi-agent-based autonomous spectral data analysis method disclosed in an embodiment of the present invention. Figure 2 This is a schematic diagram of a module of a multi-agent-based autonomous spectral data analysis device disclosed in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention.

[0017] Explanation of reference numerals in the attached drawings: 201, acquisition module; 202, processing module; 203, output module; 301, processor; 302, communication bus; 303, user interface; 304, network interface; 305, memory. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification 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.

[0019] In the description of the embodiments of the present invention, words such as "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "for example" or "for instance" in the embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of words such as "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0020] In the description of the embodiments of the present invention, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0021] Existing spectral data analysis techniques mostly adopt a serial processing mode based on human experience or a single algorithm flow. Each processing stage is independent and the parameter settings are fixed. It is difficult for data quality assessment to provide continuous constraints on preprocessing, peak analysis, and phase determination. This leads to problems such as preprocessing destroying the structure of key spectral peaks, lack of consistent interpretation between peak analysis results and phase determination, in cases of complex samples, low signal-to-noise ratio, or multiple phase superposition, inability to systematically arbitrate conflicting candidate phases, and difficulty in tracing and verifying analytical conclusions. Overall, it is difficult to simultaneously meet the comprehensive requirements of analytical stability, accuracy, and interpretability.

[0022] This embodiment discloses a multi-agent-based autonomous analysis method for spectral data, referring to... Figure 1 This includes the following steps S110-S170: S110: Obtain the original spectral data corresponding to the target analysis object and bind a preset identifier to form a set of original spectral data.

[0023] This invention discloses a multi-agent-based autonomous spectral data analysis method applied to a server. The server includes, but is not limited to, electronic devices such as mobile phones, tablets, wearable devices, and PCs (Personal Computers), and can also be a backend server running the multi-agent-based autonomous spectral data analysis method. The server can be implemented using a standalone server or a server cluster composed of multiple servers.

[0024] The quality diagnostic agent performs data quality assessment on the raw spectral dataset. Its internal structure includes a quality feature parsing unit, a quality context management unit, and a quality assessment fusion unit. The quality feature parsing unit extracts signal-to-noise characteristics, baseline stability characteristics, spike noise characteristics, spectral line saturation characteristics, and spectral axis consistency characteristics from the raw spectral data. All quality features are extracted under quality context constraints bound to preset identifiers. The quality context management unit maintains the association between acquisition condition identifiers, instrument status identifiers, and time identifiers and their corresponding raw spectral data, ensuring the comparability of quality feature parsing results across different datasets. The quality assessment fusion unit performs unified quantification and weighted fusion of various quality features, generating a data quality assessment result corresponding one-to-one with each raw spectral data point. This data quality assessment result is then written back and bound to the raw spectral dataset to form a quality-labeled spectral dataset.

[0025] The central coordinating agent is responsible for global scheduling and decision arbitration throughout the autonomous spectral data analysis process. Its internal structure includes a task decomposition unit, a constraint generation unit, a strategy scheduling unit, and an arbitration decision unit. The task decomposition unit breaks down the overall analysis task based on data quality assessment results, mapping the analysis process into multiple sub-tasks related to data quality. The constraint generation unit generates preprocessing constraints and processing intensity boundaries based on the data quality assessment results, limiting the impact of subsequent agents on the spectral structure during execution. The strategy scheduling unit integrates the task decomposition results and constraint information to generate a set of preprocessing strategies corresponding to the quality-labeled spectral data set, and uniformly manages the execution order and applicable scope of each strategy. The arbitration decision unit performs consistency checks and conflict arbitration on the spectral peak parameter set and phase inference result set during the phase inference stage, thereby outputting the final analysis result set.

[0026] The peak-fitting agent performs peak analysis on a preprocessed spectral dataset. Its internal structure includes a peak detection unit, a parameter initialization unit, a fitting optimization unit, and a result verification unit. The peak detection unit performs local extremum and morphological feature analysis on the preprocessed spectral data to obtain candidate peak positions while maintaining the spectral axis index mapping. The parameter initialization unit generates initial peak parameters based on the candidate peak positions and limits the value ranges of peak position, peak width, peak shape, and peak intensity parameters. The fitting optimization unit constructs a peak superposition model and introduces robust loss constraints and complexity penalty constraints to iteratively optimize the initial peak parameters. The result verification unit verifies the residual structure and the completeness of key peak interpretations between the fitting results and the preprocessed spectral data, generating the final set of peak parameters.

[0027] The phase identification agent performs phase inference processing on a set of spectral peak parameters. Its internal structure includes a candidate template screening unit, a feature alignment unit, a matching evaluation unit, and an evidence fusion unit. The candidate template screening unit selects candidate templates from a multi-source reference spectral library that are comparable to the set of spectral peak parameters in terms of peak position distribution and spectral coverage. The feature alignment unit establishes a peak position alignment mapping relationship between the set of spectral peak parameters and the candidate templates. The matching evaluation unit calculates matching metrics such as peak position deviation, peak width matching degree, peak shape similarity, and peak intensity ratio consistency. The evidence fusion unit combines prior sample information with spectral library confidence weights to perform weighted fusion of matching results from different reference spectral libraries, generating a set of phase inference results and binding corresponding evidence chain information.

[0028] By having each of the aforementioned intelligent agents structurally undertake different functions such as quality assessment, global coordination, spectral peak analysis, and phase inference, and forming a collaborative working relationship under the unified scheduling and arbitration of the central coordinating intelligent agent, the spectral data analysis process can achieve cross-stage information transmission, consistency verification, and conflict arbitration while maintaining the correlation of data structures, thus forming a complete multi-agent autonomous spectral data analysis system.

[0029] When acquiring the original spectral data corresponding to the target analysis object and forming the original spectral data set, firstly, under the premise of clearly defining the target analysis object, a unique analysis object identifier is established for the target analysis object. The analysis object identifier is used to uniformly refer to the same analysis object throughout the entire autonomous spectral data analysis process and serves as the basic index for subsequent data association and result traceability.

[0030] After the target analysis object is identified, the spectral acquisition device is invoked to perform spectral acquisition operations on the target analysis object according to the detection requirements and experimental conditions corresponding to the target analysis object. The spectral acquisition operation is completed under the constraints of preset acquisition parameters, which include at least the spectral axis range, sampling resolution, integration time, and resampling strategy to ensure that the acquired spectral data meets the requirements of subsequent analysis in terms of structure.

[0031] After completing the spectral acquisition operation, each acquired spectral data is received and buffered as raw spectral data. During the reception process, the original correspondence between the spectral intensity sequence and the spectral axis sequence is kept unchanged to ensure the integrity of the original structure of the spectral data.

[0032] After the raw spectral data is received, a preset identifier is bound to each raw spectral data. The preset identifier includes at least an acquisition condition identifier, an instrument status identifier, a time identifier, and an analysis object identifier. The acquisition condition identifier is used to characterize the experimental environment and operating conditions during spectral acquisition, the instrument status identifier is used to characterize the working status of the spectral acquisition equipment at the time of acquisition, the time identifier is used to characterize the time and location of the spectral acquisition, and the analysis object identifier is used to establish a unique correspondence between the raw spectral data and the target analysis object.

[0033] After completing the binding of preset identifiers, a consistency verification process is performed on the original spectral data and their corresponding preset identifiers to confirm that the binding relationship between each preset identifier and the original spectral data is complete and conflict-free. After the verification is passed, all the original spectral data with completed identifier bindings are aggregated and stored according to the analysis object identifier, thereby forming a set of original spectral data that corresponds one-to-one with the target analysis object. This ensures that the original spectral data set has both spectral structure information and complete contextual identifier information at the data level, providing a unified and traceable data foundation for subsequent data quality assessment and analysis task decomposition.

[0034] S120: The quality diagnostic agent performs data quality assessment processing on the original spectral data set and generates data quality assessment results corresponding to each original spectral data.

[0035] In one possible implementation, a quality diagnostic agent performs data quality assessment processing on the raw spectral dataset to generate data quality assessment results corresponding to each raw spectral data. Specifically, this includes: establishing a quality assessment context for each raw spectral data, which at least includes acquisition condition identifiers, instrument status identifiers, and time identifiers; under the constraints of the quality assessment context, jointly modeling the global amplitude distribution, local fluctuation distribution, and noise proportion of the spectral intensity sequence for each raw spectral data to extract signal-to-noise features characterizing the level of random noise; fitting the slowly varying trend term within a preset spectral axis interval for each raw spectral data and evaluating the continuity and consistency of the fitting residual as the spectral axis position changes to generate baseline stability features characterizing the degree of baseline drift; and analyzing the phase... The intensity abrupt change relationship between adjacent spectral axis sampling points and the probability of repeated occurrence of abrupt change points within a local window are used to extract spike noise features to characterize isolated spikes and outliers. For each raw spectral data, spectral saturation features are generated to characterize the risk of effective information truncation by detecting whether the spectral intensity reaches the upper limit of the instrument's range within the continuous spectral axis interval. Based on the alignment relationship between multiple raw spectral data in the raw spectral dataset, spectral axis consistency features are generated to characterize the stability of the spectral axis by analyzing the relative positional deviation of characteristic peaks in different raw spectral data. Signal-to-noise ratio features, baseline stability features, spike noise features, spectral line saturation features, and spectral axis consistency features are mapped to a unified evaluation scale and assigned corresponding weight parameters according to a preset quality evaluation strategy to generate data quality evaluation results corresponding to each raw spectral data.

[0036] Specifically, when establishing a quality assessment context for each raw spectral data point, the acquisition condition identifier, instrument status identifier, and time identifier, each uniquely bound to the raw spectral data, are first unpacked from the preset identifier. These three identifiers, along with the spectral data identifier of the raw spectral data, are then written into the same context record. Simultaneously, the context record mirrors the acquisition parameters that vary with acquisition conditions, such as spectral axis range, sampling resolution, integration time, and resampling strategy, to ensure that various extracted quality features can be compared horizontally under the same discrimination benchmark. The acquisition condition identifier characterizes the set of environmental and operational conditions during acquisition, such as temperature control status, optical path configuration status, sample position status, and background subtraction status. The instrument status identifier characterizes the set of instrument operating states at the time of acquisition, such as light source status, detector gain status, dark current compensation status, and range configuration status. The time identifier characterizes the time position of the acquisition and is used to establish time series relationships between multiple raw spectral data points under the same analytical object identifier, thereby supporting subsequent alignment relationship construction and stability discrimination.

[0037] When extracting signal-to-noise (SNR) features under the constraints of quality assessment context, for each original spectral data, the spectral intensity sequence is divided into several local windows according to the spectral axis position. The amplitude distribution and dynamic range of the intensity sequence are statistically analyzed at the global scale, while the fluctuation distribution and high-frequency residual proportion within the local windows are statistically analyzed at the local scale. This separates random noise from structural spectral peaks and forms quantifiable indicators. The global amplitude distribution is used to characterize the overall energy level and extreme value distribution, the local fluctuation distribution is used to characterize the short-scale fluctuation intensity, and the noise proportion is used to characterize the proportion of high-frequency components in the intensity sequence that are difficult to explain by the smooth structure. The expression for the SNR feature is: in, This represents the signal-to-noise ratio value; the larger the value, the weaker the random noise. The signal power estimate is obtained by selecting a set of spectral bands containing stable spectral peak structures from the original spectral data and calculating the energy statistics of the intensity sequence within that set of spectral bands. This represents the noise power estimate, obtained by smoothing and fitting the same original spectral data to obtain the structure term, then subtracting the structure term from the original intensity sequence to obtain the high-frequency residual sequence, and finally statistically analyzing the energy of this high-frequency residual sequence. This expression ensures that the original spectral data at different intensity scales remain comparable by ratioizing the energy of the structure term to the energy of the high-frequency residual, and enables the subsequent preprocessing intensity boundary to adaptively converge with the random noise level.

[0038] When generating baseline stability features, the spectral axis range and available spectral band constraints are first read in the quality assessment context. Then, a slow-varying trend term is fitted within a preset spectral axis interval. This slow-varying trend term describes low-frequency background changes caused by non-peaks, including source drift, fluorescence background, scattering background, or system bias. After fitting, the continuity and consistency of the fitting residuals along the spectral axis are calculated. Continuity is used to determine if there is a long-segment bias in the residuals, and consistency is used to determine if the statistical characteristics of the residuals change systematically with the spectral axis position, thereby quantifying the risk of baseline drift. The expression for the baseline stability feature is: in, The value represents the baseline stability characteristic; a larger value indicates poorer baseline stability. N represents the number of sampling points within the preset spectral axis interval. This represents the fitting residual for the i-th sampling point. The fitting residual is obtained by subtracting the fitted value of the slow-varying trend term from the observation intensity of that sampling point. This represents the fitting residual of the adjacent previous sampling point. This expression characterizes the intensity of the residual variation along the spectral axis by accumulating the absolute values ​​of the differences between adjacent residuals. When the slowly varying trend term is underfitted or the baseline drifts, the residual will exhibit a continuous shift or slow fluctuations, thus pushing up the spectral density. This reflects baseline instability.

[0039] When extracting peak noise features, a difference sequence is first formed for adjacent sampling points using the spectral axis index relationship. Candidate labels are then assigned to abrupt change locations within this difference sequence. The abrupt change relationship characterizes the degree of discontinuous intensity jumps between adjacent spectral axis sampling points. Subsequently, a local window is constructed centered on the abrupt change location. The probability of recurrence and the degree of amplitude anomaly of the abrupt change point within the local window are statistically analyzed to distinguish isolated peaks from the edges of true spectral peaks. Isolated peaks typically exhibit anomalous elevation or collapse at a single point or a few points and lack the continuous shape of a spectral peak. The expression for peak noise features is: Where S represents the value of the spike noise feature, and the larger the value, the more significant the spike noise; M represents the number of candidate mutation locations detected. This represents the spectral axis index corresponding to the k-th candidate mutation position; Indicating in the index The adjacent difference value at a given index is obtained by the difference between the intensity at that index and the adjacent intensity. This represents the mutation threshold, the value of which is determined jointly by the noise level in the quality assessment context and the preset sensitivity strategy. This indicates an indicator function that takes the value 1 when the condition is met and 0 otherwise. This expression accumulates and normalizes the amplitude of mutations exceeding a threshold, ensuring that spike noise is affected by both the frequency of occurrence and the amplitude intensity. This facilitates the subsequent spike repair subtask in adaptively selecting the repair range and intensity within the processing intensity boundary.

[0040] When generating spectral line saturation features, first read the upper limit configuration from the instrument status indicator, and then scan the raw spectral data for flat-topped or truncated segments within the continuous spectral axis interval where the intensity reaches or approaches the upper limit. Simultaneously record the length, frequency, and location of the truncated segments. Spectral line saturation characterizes the state where the detector output can no longer change with the increase of the true signal, which can easily lead to the truncation of spectral peaks and introduce phase inference bias. The expression for the spectral line saturation feature is: Where H represents the value of the spectral saturation characteristic, and the larger the value, the higher the risk of saturation; N represents the number of sampling points; This represents the observation intensity at the i-th sampling point; Indicates the upper limit of the range given by the instrument status indicator; This represents the saturation neighborhood threshold, used to define the discrimination bandwidth approaching the upper limit. Its value is determined by the noise level and the instrument's quantization resolution strategy. This expression measures the truncation risk by the percentage of sampling points whose statistical intensity falls within the upper limit neighborhood. When a flat top or truncation occurs, the number of sampling points in the upper limit neighborhood increases significantly, causing H to rise and driving subsequent protective strategies or triggering reacquisition strategies for the saturation segment.

[0041] When generating spectral axis consistency features, an alignment relationship is first established for the original spectral data set under the same analytical object identifier. This alignment relationship describes the correspondence between different original spectral data in the spectral axis coordinate system. It can be established directly based on instrument calibration information or based on calibration using stable characteristic peaks. Subsequently, under the constraint of the alignment relationship, the relative positions of corresponding characteristic peaks are extracted from multiple original spectral data, and their relative positional deviations are calculated. This quantifies the stability of the spectral axis and identifies spectral axis drift or calibration errors. The expression for the spectral axis consistency feature is: Where A represents the value of the spectral axis consistency feature, and the larger the value, the worse the spectral axis consistency; P represents the number of characteristic peaks used for alignment evaluation; This represents the peak position parameter of the p-th characteristic peak in the current original spectral data, obtained by characteristic peak detection and local fitting; This represents the reference peak position parameter of the p-th characteristic peak, calculated from other original spectral data corresponding to the same analytical object under the alignment relationship. The reference peak position parameter can be obtained using the mean, weighted mean, or robust statistics. This expression reflects the overall level of spectral axis drift by converging the characteristic peak position deviation, enabling subsequent spectral axis alignment subtasks to select the alignment model complexity and upper limit of the alignment amplitude based on the drift intensity.

[0042] When generating data quality assessment results, signal-to-noise characteristics, baseline stability characteristics, spike noise characteristics, spectral saturation characteristics, and spectral axis consistency characteristics are first mapped to a unified assessment scale. This unified assessment scale is used to convert features with different statistical calibers into a weighted, fusion-based dimensionless score. A preset quality assessment strategy is then used to assign weight parameters to each feature, expressing the different tolerances to noise, drift, saturation, and other factors across different application scenarios. The expression for the data quality assessment result is: Where Q represents the value of the data quality assessment result, which ranges from 0 to 1. The larger the value, the better the data quality. This represents a nonlinear function that compresses a linear combination into the interval between 0 and 1, used to interpret the fusion score as a quality confidence level; The scaled value representing the signal-to-noise ratio feature is obtained by linear normalization or quantile mapping of the signal-to-noise ratio feature. The scaled value represents the baseline stability feature. Inverse mapping is usually used during scaling to improve stability and correspond to a higher score. The scaled value representing the characteristics of spike noise is usually achieved using an inverse mapping during scaling. The scaled value representing the saturation characteristics of the spectral line is usually obtained by inverse mapping during scaling. The scaled value represents the spectral axis consistency feature, and the inverse mapping is usually used during scaling. to 'b' represents the weighting parameter, whose value is determined by a preset quality assessment strategy and can be calibrated based on historical analysis success rates; 'b' represents the bias parameter, used to adjust the overall quality discrimination threshold. This expression weights and fuses multidimensional quality evidence on a unified scale, and then nonlinearly compresses it to form a stable quality score. This allows the central coordinating agent to directly determine preprocessing constraints and processing intensity boundaries based on Q and the individual scaling values ​​simultaneously, thereby maintaining a strict correlation between the subsequent preprocessing strategy generation and the quality status of the original spectral data.

[0043] S130: The central coordinating agent decomposes the analysis task based on the data quality assessment results and determines the preprocessing constraints and processing intensity boundaries that match the data quality assessment results, thereby generating a set of preprocessing strategies corresponding to the quality-labeled spectral data set.

[0044] In one possible implementation, a central coordinating agent decomposes the analysis task based on the data quality assessment results and determines preprocessing constraints and processing intensity boundaries that match the data quality assessment results, thereby generating a set of preprocessing strategies corresponding to the quality-labeled spectral data set. Specifically, this includes: parsing the data quality assessment results to identify the corresponding quality feature type, anomaly degree, and repair level in each data quality assessment result, thereby generating a corresponding quality constraint description for each original spectral data set; based on the quality constraint description, performing task decomposition processing on the overall spectral analysis task, splitting the analysis task into preprocessing subtasks associated with data quality features; binding corresponding quality trigger conditions to each preprocessing subtask, so that each preprocessing subtask is activated when the corresponding quality anomaly condition is met; and determining preprocessing constraints for each preprocessing subtask based on the quality constraint description, whereby the preprocessing constraints are used to limit... The preprocessing subtasks are defined as follows: They influence the spectral structure by defining boundaries. Based on the quantification levels of each quality feature in the data quality assessment results, processing intensity boundaries are determined for each preprocessing subtask, limiting the range of parameter adjustments within the corresponding subtask. The preprocessing subtasks, preprocessing constraints, and processing intensity boundaries are jointly scheduled and arranged according to the data dependencies and structural influence order among them, generating a preprocessing decision scheme. Based on this decision scheme, the preprocessing subtasks, preprocessing constraints, and processing intensity boundaries are combined to generate a preprocessing strategy set corresponding to the quality-labeled spectral data set. Each preprocessing strategy in this set explicitly indicates the combination of preprocessing subtasks to be executed for the corresponding original spectral data and its limited parameter range. The output of this preprocessing strategy set guides subsequent spectral preprocessing.

[0045] Specifically, when parsing and processing the data quality assessment results, the central coordinating agent first reads the data quality assessment result corresponding to each original spectral data from the quality-labeled spectral data set, maintaining the binding relationship between the data quality assessment result and the original spectral data. Then, the data quality assessment result is decomposed into three types of structured fields: a set of quality feature types, a set of anomaly severity, and a set of repair levels. The quality feature types indicate the quality dimensions marked as abnormal or high-risk in the data quality assessment results, including random noise anomalies corresponding to signal-to-noise characteristics, baseline drift anomalies corresponding to baseline stability characteristics, peak anomalies corresponding to peak noise characteristics, saturation anomalies corresponding to spectral line saturation characteristics, and spectral axis drift anomalies corresponding to spectral axis consistency characteristics. The anomaly severity characterizes the deviation intensity of the corresponding quality dimension, and the repair level characterizes the feasibility and risk level of repairing the anomaly within the given preprocessing capability boundaries. To ensure that the anomaly severity and repair level can be directly used by subsequent subtask scheduling, the expression for the anomaly severity is: in, The value represents the degree of abnormality of the k-th quality characteristic, ranging from 0 to 1. The larger the value, the stronger the abnormality. This represents the scaled value of the k-th quality feature, obtained by mapping the corresponding quality feature to a unified evaluation scale. This represents the anomaly trigger threshold for the k-th quality characteristic, and its value is preset by the quality assessment strategy or obtained from historical successful cases. This represents the upper bound threshold for severe anomalies in the k-th quality characteristic, and its value is used to limit the saturation point of the anomaly degree. This represents the truncation function, used to restrict the calculation results to the range of 0 to 1. This expression normalizes the extent of the scaled value's exceedance relative to the threshold, achieving comparability of anomaly intensities across different quality dimensions. This allows the central coordinating agent to generate quality constraint descriptions with a unified standard. The quality constraint description encapsulates the quality feature type, anomaly degree, and repair level into constraint entries that correspond one-to-one with a single original spectral data point, serving as the sole basis for subsequent task decomposition, triggering conditions, and intensity boundary calculations.

[0046] When performing task decomposition based on quality constraint descriptions, the central coordinating agent maps the overall spectral analysis task into a set of preprocessing subtasks. The quality feature types in the quality constraint descriptions serve as the decomposition driving factors, ensuring that each preprocessing subtask has a clear correspondence with at least one quality feature type. Preprocessing subtasks refer to independently schedulable and constrained preprocessing functional units, including baseline correction, noise suppression, peak repair, spectral axis alignment, and intensity scale adjustment subtasks. Each preprocessing subtask carries its target object, set of adjustable parameters, and description of its structural influence range. During task decomposition, the spectral axis index mapping of the original spectral data remains unchanged. An initial dependency graph is established for the set of preprocessing subtasks according to the data dependency logic: first, eliminating global background influences; then, suppressing random perturbations; then, repairing local anomalies; then, correcting the coordinate system; and finally, unifying the intensity scale. This allows subsequent joint scheduling to generate executable preprocessing decision schemes based on the dependency graph.

[0047] When binding quality trigger conditions to each preprocessing subtask, the central coordinating agent transforms the anomaly level and repair level in the quality constraint description into decidable activation rules and writes these rules into the trigger field of the corresponding preprocessing subtask. This ensures that the preprocessing subtask only enters the executable state when the trigger field is met. The quality trigger condition is used to characterize the decision logic for whether to execute the preprocessing subtask, and includes at least two parts: an anomaly level threshold condition and a repair level allowance condition. The anomaly level threshold condition prevents minor anomalies from triggering unnecessary processing, while the repair level allowance condition prevents uncontrollable repairs from causing structural damage. The expression for the quality trigger condition is: in, This represents the activation flag for the j-th preprocessing subtask, with a value of 0 or 1. This represents the index of the quality feature type corresponding to the j-th preprocessing subtask, such as the signal-to-noise ratio feature corresponding to the noise suppression subtask and the baseline stability feature corresponding to the baseline correction subtask. This indicates the degree of abnormality of the corresponding quality characteristic type. This represents the activation threshold for the corresponding quality feature type, and its value is set by the quality assessment strategy. This indicates the repair level for the corresponding quality characteristic type, and its value is given by the quality constraint description. This represents the minimum allowed repair level threshold. This expression ensures that the activation of preprocessing subtasks is strictly consistent with the data quality status by simultaneously constraining the anomaly strength and repair feasibility, and provides a clear execution premise for subsequent strength boundary calculations.

[0048] When determining preprocessing constraints for each preprocessing subtask based on the quality constraint description, the central coordinating agent explicitly expresses the structure preservation requirement as a verifiable boundary constraint and writes this boundary constraint as a hard constraint during the execution of the preprocessing subtask into the preprocessing constraint field. Preprocessing constraints are used to limit the boundaries by which preprocessing subtasks affect the spectral structure. Spectral structure here refers to the structural elements that need to maintain a mappable relationship between the original spectral data and the preprocessed spectral data, including spectral axis index mapping, key spectral band positions, weak peak structure, and peak position stability. Preprocessing constraints include at least key spectral band preservation constraints, weak peak protection constraints, peak position drift upper limit constraints, and intensity change ratio constraints. Key spectral band preservation constraints limit the morphological changes of specific spectral bands from exceeding a threshold; weak peak protection constraints limit the peak intensity attenuation in weak peak regions from exceeding a threshold; peak position drift upper limit constraints limit any preprocessing operation from introducing peak position shifts beyond the allowable range; and intensity change ratio constraints limit the amplitude scaling range introduced by normalization or smoothing. For ease of unified execution, the expression for the peak position drift upper limit constraint is: in, This represents the peak position parameter of the p-th key spectral peak after preprocessing, which is obtained by fitting the preprocessed spectral data within the neighborhood of the key spectral peak. This represents the peak position parameter of the same key spectral peak before preprocessing, which is obtained by fitting the original spectral data within the corresponding neighborhood. This represents the maximum allowable peak position drift threshold for the p-th key spectral peak, and its value is jointly determined by the acquisition condition flag, instrument status flag, and spectral axis resolution. This constraint, by treating peak position stability as a hard boundary, ensures that preprocessing does not alter the core structural evidence upon which subsequent phase identification relies.

[0049] When determining the processing intensity boundary based on the quantification level of each quality feature in the data quality assessment results, the central coordinating agent maps the anomaly level to an allowable range of preprocessing parameters and writes this allowable range into the processing intensity boundary field. This ensures that preprocessing subtasks can only adjust parameters within the allowable range during execution. The processing intensity boundary limits the range of parameter adjustments in the corresponding preprocessing subtask. Parameter adjustments here refer to the smoothing window size, filter intensity coefficient, baseline fitting order or node density, peak repair neighborhood radius, spectral axis alignment model order or maximum displacement, normalization scaling upper limit, etc. To ensure that the intensity boundary changes continuously with the anomaly level, the expression for the processing intensity boundary is: in, This represents the target intensity parameter of the j-th preprocessing subtask. This represents the minimum allowable strength parameter for the preprocessing subtask, and its value is used to avoid invalid processing due to excessively low strength when the task is active. This represents the maximum allowable strength parameter for this preprocessing subtask, and its value is used to avoid structural damage caused by excessive strength. This indicates the degree of anomaly of the quality feature type corresponding to the preprocessing subtask. This expression linearly maps the degree of anomaly to a parameter range, so that the stronger the anomaly, the closer the processing intensity is to the upper limit, while being subject to hard boundary constraints of the preprocessing conditions, thus forming a controllable balance between repair capability and structure preservation.

[0050] When jointly scheduling preprocessing subtasks, preprocessing constraints, and processing intensity boundaries to generate preprocessing decision schemes, the central coordinating agent uses the preprocessing subtask dependency graph as the main constraint. It injects activation markers, preprocessing constraints, and processing intensity boundaries into the scheduling model and performs secondary orchestration of the execution order of preprocessing subtasks to ensure that preprocessing subtasks executed earlier do not disrupt the structural information required by later preprocessing subtasks. The core of joint scheduling is to simultaneously satisfy three types of constraints: whether to execute, to what extent to execute, and what constraints must not be disrupted during execution, and to output an executable preprocessing decision scheme for the same original spectral data. The preprocessing decision scheme describes the sequence of preprocessing subtasks to be executed for the original spectral data, the intensity parameter values ​​for each preprocessing subtask, and the set of preprocessing constraints that need to be verified in real time. It also includes insertion points for structural consistency checks between subtask executions. To quantify orchestration priority, the sorting score expression for preprocessing subtasks is: in, This represents the ranking score of the j-th preprocessing subtask; a higher score indicates that it should be executed first. This indicates the activation flag. Indicates the degree of abnormality. The structural influence coefficient is used to characterize the potential perturbation intensity of the spectral structure by the preprocessing subtask. The structural influence coefficient can be set by historical evaluation or expert strategy. and This represents a weighting coefficient used to balance prioritizing the most anomalous issues with those having less or more critical structural impacts. By incorporating both anomalous strength and structural influence into the ranking, this expression ensures that the scheduling prioritizes the elimination of critical anomalies while preventing early actions from disrupting the alignment or fit foundation for subsequent operations.

[0051] When generating a preprocessing strategy set based on the preprocessing decision scheme, the central coordinating agent reads the preprocessing decision scheme for each raw spectral data and combines and encapsulates the preprocessing subtask sequence, the preprocessing constraint set, and the processing intensity boundary set to form a preprocessing strategy entry that corresponds one-to-one with the raw spectral data. All preprocessing strategy entries are then aggregated according to the analysis object identifier to form a preprocessing strategy set. This preprocessing strategy set guides subsequent spectral preprocessing. Each preprocessing strategy explicitly indicates the combination of preprocessing subtasks to be executed, the limited parameter range of each preprocessing subtask, and the preprocessing constraints that must be continuously met during execution, thus ensuring that the preprocessing process is reproducible, traceable, and strongly correlated with the data quality status. When outputting the preprocessing strategy set, a strategy version identifier and a strategy application identifier are also output. The strategy version identifier distinguishes strategy sets generated by different scheduling iterations, and the strategy application identifier indicates which raw spectral data and its quality constraint description the strategy entry applies to. This ensures that subsequent spectral preprocessing can strictly follow the strategy and record the execution trajectory without disrupting the spectral axis index mapping relationship.

[0052] S140, based on the preprocessing strategy set, perform spectral preprocessing on the original spectral data set to obtain a preprocessed spectral data set.

[0053] In one possible implementation, spectral preprocessing is performed on the original spectral data set according to a set of preprocessing strategies to obtain a preprocessed spectral data set. Specifically, this includes: establishing a preprocessing execution context based on the set of preprocessing strategies, which defines the execution order of preprocessing subtasks; performing spectral preprocessing operations sequentially for each original spectral data set under the constraints of the preprocessing execution context; modeling and correcting the slowly varying trend term according to preprocessing constraints when performing the baseline correction preprocessing subtask; suppressing random noise components according to processing intensity boundaries when performing the noise suppression preprocessing subtask; performing local replacement or reconstruction of abnormal sampling points when performing the peak repair preprocessing subtask; performing normalization without changing the number and order of spectral axis alignment adjustment preprocessing subtask; and summarizing the processed spectral data in a predetermined order after completing each preprocessing subtask to generate a preprocessed spectral data set.

[0054] Specifically, when establishing the preprocessing execution context, the preprocessing strategy bound to each original spectral data is first read from the preprocessing strategy set. Then, a preprocessing execution record is generated for that original spectral data under the same analysis object identifier. This record preserves the preprocessing subtask sequence, the processing intensity boundaries of each subtask, the preprocessing constraints of each subtask, and the spectral axis index mapping relationship. The preprocessing execution context is used to uniformly constrain the executability of preprocessing subtasks at runtime. Its core fields include an execution order field and a dependency verification field. The execution order field indicates the sequential relationship between preprocessing subtasks, while the dependency verification field indicates the structural preservation state that each preprocessing subtask must satisfy before starting, thus ensuring that the preprocessing process can be reproducibly executed under structural consistency constraints.

[0055] During spectral preprocessing, for each piece of raw spectral data, its preprocessing execution context is first loaded and the spectral axis index mapping is locked. Then, preprocessing subtasks are activated one by one according to the execution order fields and executed sequentially on the same data stream. The execution order is achieved by assigning a stage identifier to each preprocessing subtask and executing the processing in ascending order of stage identifiers. Simultaneously, after each stage, the dependency verification fields are validated to confirm that the stage has not disrupted the positions of key spectral segments, weak peak structures, and peak position drift upper limits. Spectral preprocessing here refers to constrained adjustments to the spectral intensity sequence to improve analysibility while maintaining the spectral axis sequence and spectral axis index mapping relationship unchanged, thereby ensuring that the subsequent peak fitting agent can reuse the structural position mapping relationship within the same spectral axis coordinate system.

[0056] When performing the baseline correction preprocessing subtask, the modeling spectral segment and modeling complexity of the slow-varying trend term are first determined under the constraints of the preprocessing execution context. Then, the spectral intensity sequence is fitted with the slow-varying trend term while maintaining the spectral axis index mapping relationship. The slow-varying trend term characterizes low-frequency background changes caused by source drift, background radiation, or system bias. Its fitting employs a smoothing constraint consistent with the preprocessing constraints to avoid absorbing the true spectral peak shapes into the slow-varying trend term. After completing the slow-varying trend term fitting, the baseline correction result is obtained by subtracting the fitted value of the slow-varying trend term from the original spectral intensity sequence. The upper limit constraint on peak position drift is immediately verified. If the upper limit constraint on peak position drift is triggered, the modeling complexity is reduced or the modeling spectral segment range is narrowed to ensure that the baseline correction does not change the structural position mapping relationship of key spectral peaks.

[0057] When executing the noise suppression preprocessing subtask, the processing intensity boundary is read under the constraints of the preprocessing execution context and mapped to a set of noise suppression parameters. This set of parameters includes at least the smoothing window size, filter intensity coefficients, and a local adaptive threshold. Random noise components refer to perturbation components that exhibit high-frequency, irregular fluctuations along the spectral axis and lack continuous peak shapes. Constrained suppression means that the noise suppression parameters can only be adjusted within the range defined by the processing intensity boundary, and the noise suppression process must satisfy weak peak protection constraints and key spectral segment preservation constraints. After noise suppression is executed, the local curvature changes at the key spectral segment positions are verified. If signs of weak peak structures being suppressed by smoothing are observed, the smoothing window size is reduced downwards, and noise suppression is re-executed to ensure consistency between noise reduction and structure preservation.

[0058] When performing the peak repair preprocessing subtask, under the constraints of the preprocessing execution context, the set of anomalous sampling points is first located. Anomalous sampling points refer to sampling points whose intensity abrupt changes satisfy the peak noise discrimination condition and do not conform to the continuous peak shape. Local replacement refers to generating substitute values ​​and replacing the intensity of anomalous sampling points using neighborhood interpolation or neighborhood fitting within the neighborhood of the anomalous sampling points. Local reconstruction refers to establishing a local continuous model within the neighborhood of the anomalous sampling points and replacing the anomalous segments within the neighborhood with the intensity sequence output by the local continuous model. When performing peak repair, the radius of the repair neighborhood and the replacement amplitude are strictly limited to not exceeding the processing intensity boundary. After repair, the continuity of the neighborhood of the anomalous sampling points is verified. Continuity here means that no new abrupt peaks appear in the difference between adjacent sampling points, thereby avoiding the introduction of new artifact structures by peak repair and maintaining the stability of the structural position mapping relationship.

[0059] When performing the spectral axis alignment and adjustment preprocessing subtask, the number and order of spectral axis sampling points are locked under the constraints of the preprocessing execution context. This ensures that any alignment operation cannot change the length and order structure of the spectral axis index mapping relationship. Then, based on the alignment model, restricted resampling or restricted shift compensation is performed on the spectral intensity sequence to make different original spectral data comparable under a unified spectral axis reference. Spectral axis alignment refers to eliminating the characteristic peak position deviation caused by spectral axis drift. Normalization refers to scaling the spectral intensity sequence to eliminate amplitude differences caused by different acquisition conditions while keeping the number and order of spectral axis sampling points unchanged. This stage is simultaneously constrained by the upper limit of peak position drift and the proportional constraint of intensity change, ensuring that alignment and normalization do not lead to uncontrollable changes in the spectral peak position mapping relationship or relative intensity structure. After processing, the characteristic peak alignment residuals are verified to confirm that the spectral axis consistency has been improved.

[0060] When generating the preprocessed spectral data set, under the constraints of the preprocessing execution context, the results of baseline correction, noise suppression, peak repair, and spectral axis alignment adjustments for each original spectral data are written into the preprocessed spectral data set in a predetermined order. The spectral axis index mapping relationship, key spectral segment position mapping relationship, preprocessing subtask sequence, and parameter value records for each preprocessed spectral data are also fixed for each original spectral data. The predetermined order refers to the data arrangement order consistent with the original spectral data set, thus ensuring a one-to-one correspondence between the preprocessed and original spectral data sets and allowing traceability based on the analysis object identifier. This provides a stable and structurally consistent input foundation for subsequent peak analysis processing.

[0061] S150, based on the preprocessed spectral data set, the peak fitting agent performs peak analysis processing on each preprocessed spectral data to generate a set of peak parameters corresponding to each preprocessed spectral data.

[0062] In one possible implementation, based on a preprocessed spectral data set, a peak fitting agent performs peak analysis processing on each preprocessed spectral data to generate a set of peak parameters corresponding to each preprocessed spectral data. Specifically, this includes: establishing a peak analysis context based on the preprocessed spectral data set, which at least includes spectral axis index mapping relationships, key structural position mapping relationships, and preprocessing process records; under the constraints of the peak analysis context, performing initial peak detection processing on each preprocessed spectral data, generating a set of potential peak candidate positions by jointly analyzing the local extremum features, gradient change features, and curvature change features of the spectral intensity along the spectral axis; and performing adaptive peak number determination processing on each preprocessed spectral data based on the spectral morphology information in the peak analysis context, analyzing the spacing and overlap between the peak candidate positions in the peak candidate position set. The system determines the number of target peaks corresponding to each preprocessed spectral data, along with local signal-to-noise levels. Initial peak parameters are generated for each target peak, including peak position, peak width, peak shape, and peak intensity. The range of these initial peak parameters is jointly limited by the structure preservation constraints in the peak analysis context and the preprocessing records. For each preprocessed spectral data, an objective function containing a peak superposition model and a background term is constructed, and robust loss constraints are introduced during the fitting process to iteratively adjust the initial peak parameters, outputting the fitting results. The fitting results are compared with the preprocessed spectral data in terms of overall contour, local residual distribution, and key structural positions. When the peak analysis results meet the preset reliability conditions, the final peak parameters are output. Finally, a set of peak parameters corresponding to each preprocessed spectral data is generated for the final peak parameters.

[0063] Specifically, when establishing the peak analysis context, the peak fitting agent first reads the preprocessed spectral data set under the same analysis object identifier, maintaining the spectral axis index mapping relationship between each preprocessed spectral data and its corresponding original spectral data unchanged. Then, a context record is constructed for each preprocessed spectral data, and the spectral axis index mapping relationship, key structure position mapping relationship, and preprocessing process record are solidified. The spectral axis index mapping relationship is used to ensure the traceability of peak position parameters at different processing stages, the key structure position mapping relationship is used to solidify the index range of key spectral segments and weak peak structures, and the preprocessing process record is used to solidify the parameter values ​​and execution order of baseline correction, noise suppression, peak repair, and spectral axis alignment adjustment. Thus, in the subsequent detection, peak determination, fitting, and verification processes, structure preservation constraints are explicitly injected into the threshold selection and parameter boundaries of each step.

[0064] During the initial peak detection process, under the constraints of the peak analysis context, local extremum features, gradient change features, and curvature change features are calculated along the spectral axis for each preprocessed spectral data. A set of potential peak candidate positions is generated within the effective spectral range defined by the key structural position mapping relationship. Local extremum features are used to mark the maxima positions of the intensity sequence within the local window; gradient change features are used to mark the inflection points of intensity increases and decreases; and curvature change features are used to mark the morphological change positions at the peak tip and shoulder. To ensure the stability of the candidate position set, gradient sign change is used as the initial screening condition, followed by a curvature amplitude threshold as a morphological constraint. Finally, the candidate position set is merged with minimum spacing to avoid duplicate marking of the same peak. A unified discriminant can be used for candidate position discrimination, the expression of which is: in, Indicates whether index i is determined to be a candidate position of spectral peak, with a value of 0 or 1. This represents the gradient estimate at index i, obtained by adjacent intensity difference or local linear fitting. Used to express the local maximum condition where the gradient changes from positive to negative. Indicating in the index The curvature estimate at that point is obtained by second-order difference or local quadratic fitting. The curvature threshold is determined by the local signal-to-noise level in the spectral peak analysis context and the preprocessing records. This expression suppresses spurious peaks caused by residual noise from entering the candidate position set by simultaneously treating the maximum value condition and the morphological sharpness condition as necessary conditions.

[0065] When performing adaptive peak number determination, the set of potential peak candidate positions is aggregated and evaluated under the constraints of the peak analysis context. The target peak number is determined by analyzing the spacing relationship, overlap degree, and local signal-to-noise level between candidate positions. The spacing relationship describes the distance between adjacent candidate positions on the spectral axis index; the overlap degree describes whether there is significant overlap in the local peak shapes corresponding to adjacent candidate positions; and the local signal-to-noise level describes the significance of the peak structure in the neighborhood of the candidate position relative to the residual noise. To unify the merging of overly dense candidate positions and the splitting of overlapping peaks to the same discrimination scale, the resolvability is first calculated for each pair of adjacent candidate positions, and then a decision on whether to merge is made based on the resolvability threshold, thus obtaining the target peak number. The expression for resolvability is: Where D(p,q) represents the distinguishability between candidate position p and candidate position q, and the larger the value, the easier it is to distinguish them into two spectral peaks. and This represents the initial peak estimate corresponding to the candidate position, which is obtained from the spectral coordinates of the candidate position or through local fitting. and This represents the initial estimate of the peak width corresponding to the candidate position, which is obtained by estimating the full width at half maximum (FWHM) or by local fitting within the neighborhood of the candidate position. This expression normalizes the candidate position spacing to the average peak width scale, ensuring consistency in the merging discrimination under different resolutions or peak widths, thereby obtaining the number of target spectral peaks consistent with the shape of the preprocessed spectral data.

[0066] When generating initial spectral peak parameters, for each target spectral peak with a determined number of target peaks, peak position parameters, peak width parameters, peak shape parameters, and peak intensity parameters are generated under the constraints of the spectral peak parsing context. The value range of each type of parameter is written into the parameter boundary field of the target spectral peak. The peak position parameter describes the center position of the spectral peak, the peak width parameter describes the extent of the spectral peak's expansion along the spectral axis, the peak shape parameter describes the shape category or mixing ratio of the spectral peak, and the peak intensity parameter describes the amplitude or area intensity of the spectral peak. The parameter boundary field is jointly limited by the structure preservation constraint and the preprocessing record. The structure preservation constraint restricts the peak position parameter from deviating from the allowable range given by the key structural position mapping relationship, while the preprocessing record restricts the peak width and peak shape parameters from falling into the morphological range that has been significantly changed by the preprocessing operation, thereby avoiding conflicts between the initial value settings and the actual spectral peak morphology. The peak position parameter boundary can be given according to the key structural position mapping relationship, and the expression of the boundary is: in, This represents the peak position parameter of the p-th target spectral peak. This represents the initial estimate of the peak position obtained from the candidate positions. This indicates the allowable offset range, and its value is jointly determined by the resolution of the spectral axis index mapping, the protection strength of the key structural position mapping, and the preprocessing record. This boundary restricts the peak position search to a traceable structural neighborhood, ensuring that subsequent iterative fitting does not drift to incoherent spectral bands and maintains the structural correlation with the original spectral data.

[0067] When outputting the fitting results, an objective function containing a peak superposition model and a background term is constructed for each preprocessed spectral data. Iterative optimization is then performed using the initial peak parameters as initial values ​​under the constraints of the peak analysis context. Robust loss constraints are introduced during the iteration process to reduce the interference of outliers and local residual noise on parameter convergence. The peak superposition model represents the observed spectrum as a sum of multiple parameterized peak functions. The background term describes the slowly changing background that may still remain. The robust loss constraint reduces the influence of this point on the optimization direction when the residual is large, thereby avoiding parameter divergence caused by peak residue or local anomalies. The expression for the objective function is: in, This represents the set of parameters to be optimized, containing the peak position parameters of all target spectral peaks. Peak width parameter Peak shape parameters With peak intensity parameters And the background parameter. N represents the number of sampling points. Indicates index The preprocessed spectral intensity value at the location. This represents the spectral axis coordinate value at index i. P represents the number of target spectral peaks. This represents the peak intensity parameter of the p-th spectral peak. The peak function is represented by the peak position parameter, peak width parameter, and peak shape parameter, which together determine the peak shape. Indicates the background term in the spectral coordinates The value at that location. This represents the robust loss function, used to reduce the weight of large residual points. The objective function minimizes the robust residual within the parameterized peak model framework, ensuring that the peak parameters converge stably even in the presence of noise and local anomalies. Furthermore, parameter updates are always constrained by the aforementioned parameter boundary fields, thus yielding interpretable fitting results.

[0068] When the preset reliability conditions are met and the final spectral peak parameters are output, a multi-layer consistency check is performed on the fitted spectral peak model and the preprocessed spectral data under the constraints of the spectral peak analysis context. This includes overall contour matching check, local residual distribution check, and key structure position matching check. The overall contour matching check checks the consistency of the fitted curve and the observed curve across the entire spectrum. The local residual distribution check checks whether the residuals exhibit an unstructured random distribution and avoids continuous deviation bands. The key structure position matching check checks whether the residuals within the spectral band defined by the key structure position mapping relationship are below a threshold and whether the key spectral peaks are fully interpreted. The preset reliability conditions unify the above check results into a definite pass condition, and its expression is: Where R represents the reliability determination result, and its value is 0 or 1. Indicates index The residual at the point is obtained by subtracting the output of the fitted model from the observation intensity. This represents the global mean square error threshold, the value of which is determined jointly by the local signal-to-noise level and the records from the preprocessing process. A set of key indexes that defines the mapping relationship between key structural locations. This represents the maximum residual threshold within the key index set, and its value is used to strengthen the structural preservation requirements of key spectral segments. This determination avoids situations where the global error is small but the interpretation of key spectral segments fails, by simultaneously constraining both the global fitting error and the key spectral segment fitting error. When the value is 1, the current parameter is locked and the final spectral peak parameter is output. When the value is 0, a rollback is triggered and the process of adaptively determining the number of target spectral peaks or generating initial spectral peak parameters is returned to perform correction.

[0069] When generating the spectral peak parameter set, after each preprocessed spectral data point passes reliability assessment, its final spectral peak parameters are encapsulated by spectral peak index. These parameters, along with the spectral axis index mapping relationship and the key structural position mapping relationship, are written into the spectral peak parameter set entry. Simultaneously, the spectral peak parameter set entry is bound to the preprocessing process record of the preprocessed spectral data to ensure end-to-end traceability. The spectral peak parameter set serves as the sole spectral peak evidence carrier in the subsequent phase identification agent. It contains at least the peak position parameter, peak width parameter, peak shape parameter, and peak intensity parameter for each spectral peak, maintaining a one-to-one correspondence with the preprocessed spectral data set. This allows the phase identification stage to directly reuse structural association information and perform template matching and evidence fusion without reinterpreting the preprocessing process.

[0070] S160: The phase identification agent matches the set of spectral peak parameters with the template information in the multi-source reference spectral library, and combines the sample prior information with the confidence weight of the spectral library to generate a set of phase inference results corresponding to each candidate.

[0071] In one possible implementation, a phase identification agent matches the set of spectral peak parameters with template information in a multi-source reference spectral library, and combines prior sample information with library confidence weights to generate a set of phase inference results corresponding to each candidate. Specifically, this includes: screening a set of candidate templates that are structurally comparable to the set of spectral peak parameters based on the distribution range of peak positions, peak quantity characteristics, and spectral band coverage of the multi-source reference spectral library; establishing a correspondence between the peak positions in the set of spectral peak parameters and the characteristic peak positions of the candidate templates in the set of candidate templates to form a peak alignment mapping relationship; generating a matching measurement result for each candidate template by jointly calculating peak position deviation, peak width matching degree, peak shape similarity, and peak intensity ratio consistency based on the peak alignment mapping relationship; introducing prior sample information to constrain and correct the matching measurement result, and obtaining a first template by suppressing or eliminating candidate templates that do not conform to the prior sample information; performing weighted fusion processing on the first template using library confidence weights to obtain a second template; and performing comprehensive inference processing on the candidate phases corresponding to the second template to generate phase inference results corresponding to each candidate, forming a set of phase inference results.

[0072] Specifically, when screening the candidate template set, the phase identification agent first reads the peak position parameter distribution range, peak quantity characteristics, and spectral band coverage relationship from the peak position parameter set, while maintaining the one-to-one correspondence between the peak parameter set and the preprocessed spectral data set. These three elements are then encapsulated into a retrieval constraint vector, and a multi-source reference spectral library is searched in parallel. The multi-source reference spectral library refers to a set of reference templates from different data sources or measurement conditions. Each reference template is bound to a template identifier, template spectral band range, template characteristic peak set, and template metadata. The peak position parameter distribution range characterizes the minimum and maximum value ranges of all peak position parameters in the peak parameter set. The peak quantity characteristics characterize the number of peaks in the peak parameter set and their density distribution in different spectral bands. The spectral band coverage relationship characterizes the coverage of the effective spectral band set involved in the peak parameter set and the mapping relationship between the key structural positions. During the screening process, reference templates that do not have the same spectral coverage are first filtered by spectral coverage relationship. Then, reference templates with completely non-overlapping peak positions are filtered by peak position parameter distribution range. Finally, reference templates with obviously mismatched peak densities are filtered by peak number characteristics. This results in a set of candidate templates that are comparable in terms of structural features. The library identifier of each candidate template in the set is recorded to support subsequent library confidence weight fusion.

[0073] When establishing peak alignment mapping, the phase identification agent reads the characteristic peak set of each candidate template and performs bidirectional matching between the peak position parameters in the spectral peak parameter set and the characteristic peak positions in the candidate template's characteristic peak set to establish a one-to-one correspondence. The characteristic peak position refers to the coordinates of the labeled or stably detectable peaks in the candidate template, and the peak alignment mapping is used to solidify the peak position in the spectral peak parameter set. The peak position parameter corresponds to the first peak in the candidate template. This method maps the positions of characteristic peaks to provide a unified indexing framework for subsequent metrics such as peak position deviation and peak width matching. To avoid ambiguity caused by many-to-one or one-to-many relationships, candidate matching pairs are first generated in ascending order of peak position difference. Then, matching pairs are prioritized within the key spectral segments defined by the mapping relationship of key structural positions. Finally, conflict resolution is performed on the remaining matching pairs to ensure that the mapping relationship is injective or partially injective, and missing markers are recorded for unmatched peak position parameters for inclusion in the penalty during comprehensive inference. The construction of the peak alignment mapping relationship adopts the objective of minimizing the total peak position difference, and the expression of the objective is: in, This indicates the peak alignment mapping relationship, mapping the spectral peak index p to the candidate template feature peak index or to an empty mapping. This represents the p-th peak position parameter in the set of spectral peak parameters. Indicates the candidate template with The location of the matched characteristic peak. This represents the missing marker indicator function, which takes a value of 1 when the spectral peak does not match any template feature peak. This represents the missing value penalty coefficient, which is used to balance the mismatches caused by forced matching and the information loss caused by allowing missing values. The objective is to minimize the total peak position difference while penalizing missing mappings, so that the peak position alignment mapping relationship tends to have a high consistency match, and remains robust when peaks are missing or templates are incomplete.

[0074] When generating matching metric results, the phase identification agent calculates peak position deviation, peak width matching degree, peak shape similarity, and peak intensity ratio consistency for each candidate template under the constraint of peak position alignment mapping relationship, and integrates the four metrics into a candidate template matching metric result. Peak position deviation characterizes the degree of offset between the spectral peak parameter set and the candidate template in peak position coordinates; peak width matching degree characterizes the compatibility between the peak width parameters of the spectral peak parameter set and the reference peak width of the candidate template; peak shape similarity characterizes the consistency between the peak shape parameters of the spectral peak parameter set and the peak shape category or peak shape mixing ratio of the candidate template; peak intensity ratio consistency characterizes the consistency between the relative ratio of peak intensity parameters within the spectral peak parameter set and the reference intensity ratio within the candidate template. The matching metric result serves as the main evidence for candidate phase ranking and subsequent constraint correction, and its expression is: in, This represents the matching metric result of the candidate template; a larger value indicates a better match. This represents the fusion weight of the four metrics. The value is determined by the application scenario strategy or historical case calibration, and the weight sum is 1. The peak position deviation score is obtained by normalizing the aligned peak position difference through a threshold and then mapping it inversely. The peak width matching score is obtained by normalizing the aligned peak width difference through a threshold and then mapping it inversely. The peak shape similarity score is obtained by similarity mapping based on the differences in aligned peak shape parameters. The peak intensity ratio consistency score is obtained by mapping the difference in peak intensity ratios after alignment using similarity mapping. This expression weights and converges multidimensional matching evidence into a single score, enabling subsequent sample prior information and spectral library confidence weights to consistently constrain, correct, and fuse the unified score.

[0075] When introducing prior sample information for constraint correction, the phase identification agent first parses the prior sample information and transforms it into a set of decidable prior constraints. The prior sample information describes the prior limitations of the target analyte in terms of experimental conditions, composition range, process origin, or known coexistence relationships. The set of prior constraints includes at least necessary existence constraints, necessary non-existence constraints, and conditional existence constraints. Necessary existence constraints are used to prioritize candidate templates that conform to the prior information; necessary non-existence constraints are used to directly eliminate candidate templates that contradict the prior information; and conditional existence constraints allow candidate templates to enter subsequent fusion only when evidence for a specific spectral band is met. Constraint correction is achieved by applying a suppression factor or elimination operation to the matching metric results of candidate templates, and the set of candidate templates retained after correction is defined as the first template. Suppression and elimination use a unified discriminant, the expression of which is: in, This indicates the matching metric result after constraint correction. This indicates the candidate phase identifier corresponding to the candidate template. This represents the set of permissible phases derived from prior information about the sample, when... When a phase does not belong to the set of allowed phases, it is removed using an indicator function. This represents the prior violation degree of the candidate phase. The larger the value, the stronger the conflict with the prior information of the sample. It is calculated by comparing the attribute labels of the candidate phase with the prior constraint set one by one. This represents the suppression intensity coefficient, used to control the penalty applied to the matching metric results by prior violations. This expression ensures that the first template strictly satisfies the constraints of the sample's prior information while preserving spectral peak evidence by directly eliminating disallowed phases and exponentially penalizing weakly conflicting phases.

[0076] When introducing library confidence weights for weighted fusion, the phase identification agent reads the library identifier for each candidate template in the first template, obtains the confidence weight corresponding to that library identifier from the library confidence weights, corrects the constraints of the candidate template, and performs weighted convergence on the matching measurement results. The library confidence weights characterize the reliability of different reference spectral libraries in terms of data source reliability, template completeness, and applicable measurement conditions. A higher confidence weight indicates that the template evidence provided by that library is more trustworthy. The goal of weighted fusion is to improve the evidence stability when multiple spectral libraries simultaneously support the same candidate phase, and to suppress the influence of low-confidence libraries when different libraries contradict each other, thus obtaining the second template. The expression for weighted fusion is: in, Indicates candidate phase The overall matching score after weighted fusion. L represents the number of spectral libraries. This represents the confidence weight of the l-th spectral library. This indicates that the candidate phase in the first template comes from the l-th spectral library. The set of candidate templates. This represents the matching metric result after constraint correction for candidate template t. This expression ensures that high-confidence spectral libraries contribute more to the overall matching score by weighted averaging of evidence from multiple spectral libraries at the candidate phase granularity, and transforms the support of multiple templates for the same phase into more stable comprehensive evidence, thereby forming a unified ranking basis for candidate phases in the second template.

[0077] When performing comprehensive inference processing and forming a phase inference result set, the phase identification agent uses the comprehensive matching score as the core evidence for candidate phases in the second template, and combines the coverage of peak alignment mapping relationships and the interpretability of key structural position mapping relationships to generate a confidence level for the candidate phases. Subsequently, the candidate phase identifier, confidence level, supporting template identifier set, peak alignment mapping relationship summary, and matching metric result summary are encapsulated into phase inference result entries, and sorted from high to low according to the comprehensive matching score to form the phase inference result set. The key to comprehensive inference processing is to downgrade candidate phases with high matching scores but incomplete interpretations, and to upgrade candidate phases with slightly lower matching scores but complete interpretation of key spectral segments and cross-spectral library consistency, thereby providing a more reliable candidate set and evidence chain for subsequent consistency verification and conflict arbitration by the central coordinating agent. The confidence level can be jointly determined by the comprehensive matching score and coverage, and the expression for the confidence level is: in, Indicates candidate phase The confidence level value is denoted by a value; a higher value indicates a higher level of confidence. This represents the overall matching score of the candidate phase. This represents the set of spectral peak indices that are successfully interpreted by the candidate phase under the peak alignment mapping relationship. This represents the set of key spectral peak indices defined by the mapping relationship of key structural locations. This expression uses the interpretation coverage of key spectral peaks as a multiplicative factor, ensuring that candidate phases must simultaneously satisfy both good overall matching and complete interpretation of key evidence to obtain a high confidence level, thus forming a set of structurally consistent and traceable phase inference results.

[0078] S170: The central coordinating agent performs consistency verification and conflict arbitration on the set of spectral peak parameters and the set of phase inference results to form the final set of analysis results, and outputs the results as structured analysis results.

[0079] In one possible implementation, a central coordinating agent performs consistency checks and conflict arbitration on the set of spectral peak parameters and the set of phase inference results to form a final set of analysis results, which is then output as structured analysis results. Specifically, this includes: constructing a corresponding arbitration context for each candidate phase, whereby the arbitration context includes phase identifiers, candidate template identifiers, matching metric results, sample prior information, spectral library confidence weights, and spectral peak subset identifiers corresponding to the phase inference results; under the constraints of the arbitration context, performing spectral peak evidence reconstruction processing on the set of spectral peak parameters, mapping it to a set of spectral peak evidence; performing consistency checks on the set of phase inference results, generating consistency check records corresponding to each candidate phase, wherein the consistency check processing includes spectral peak consistency checks for key spectral peak sets, prior consistency checks for sample prior information, and spectral library consistency checks for the stability of multi-source reference spectral libraries; and based on the consistency check records, explicitly modeling the conflict relationships between candidate phases by identifying the boundaries between candidate phases. A conflict graph structure is constructed to address the competition among key peak sets, the mutual exclusion between different key peak sets, and the incompatibility between candidate phases and prior information of the sample. Conflict arbitration is then performed based on this structure to generate an arbitration ranking of candidate phases. The arbitration process ranks candidate phases according to the order of prior decision, peak decision, and library decision. The prior decision eliminates or downgrades candidate phases that do not conform to the prior information of the sample; the peak decision prioritizes candidate phases that provide a complete interpretation of the key peak sets and have a reasonable residual distribution; and the library decision suppresses candidate phases with unstable cross-library support. Based on the arbitration ranking, a backtesting process is performed on a pre-set number of candidate phases. This involves remapping the corresponding candidate templates to the spectral axis index mapping of the preprocessed spectral data set and verifying whether the interpretation residuals at the key peak sets meet pre-set confidence conditions. After successful backtesting, the phase inference results that passed arbitration are structurally encapsulated to form the final analysis result set.

[0080] Specifically, when constructing the arbitration context, the central coordinating agent, while maintaining the evidence chain binding relationship between the phase inference result set and the spectral peak parameter set, generates an independent arbitration context record for each candidate phase. The corresponding phase identifier, candidate template identifier set, matching metric result, sample prior information, spectral library confidence weight, and spectral peak subset identifier are written into the same record. The arbitration context provides a unified data view for subsequent consistency verification and conflict arbitration processes. The phase identifier uniquely identifies the candidate phase; the candidate template identifier traces which reference templates support the phase; the matching metric result quantifies the matching strength between the spectral peak parameter set and the reference templates; the sample prior information limits the reasonable range of the phase's existence; the spectral library confidence weight reflects the credibility of evidence from different reference spectral libraries; and the spectral peak subset identifier indicates the set of spectral peaks that the candidate phase primarily relies on for interpretation. This ensures that the arbitration decision is traceable and interpretable at the evidentiary level.

[0081] During the spectral peak evidence reconstruction process under the constraints of the arbitration context, the central coordinating agent reads peak position parameters, peak width parameters, peak shape parameters, and peak intensity parameters from the spectral peak parameter set. Combining this with the spectral axis index mapping relationship and the key structure position mapping relationship, the parameterized spectral peak representation is converted into a unified spectral peak evidence set. This spectral peak evidence set is used in the arbitration stage to describe the existence, importance, and interpretability of spectral peaks in a unified format. Each piece of spectral peak evidence includes at least a spectral peak index, peak position interval, peak intensity weight, and a key marker, which indicates whether the spectral peak belongs to the key spectral peak set. Through spectral peak evidence reconstruction, continuous parameter space information is mapped into discrete and comparable evidence units, enabling subsequent consistency checks to directly compare the interpretability of spectral peaks among candidate phases.

[0082] When performing consistency verification on the set of phase inference results, the central coordinating agent generates a consistency verification record for each candidate phase under the joint constraints of the arbitration context and the spectral peak evidence set. The consistency verification record consists of spectral peak consistency verification results, prior consistency verification results, and spectral library consistency verification results. Spectral peak consistency verification determines whether the candidate template corresponding to the candidate phase can fully explain the key spectral peak set under the peak alignment mapping relationship, and quantifies the completeness of explanation by statistically analyzing the explanation residuals and unexplained proportions corresponding to the key spectral peaks. Prior consistency verification determines whether the attribute labels, compositional characteristics, or applicable conditions of the candidate phase conflict with or are incompatible with the prior information of the sample, and grades the intensity of conflict. Spectral library consistency verification determines whether supporting evidence from different reference spectral libraries maintains a stable ranking or shows significant discrepancies after introducing spectral library confidence weights. Through the parallel generation of these three types of consistency verification results, each candidate phase receives a clear evaluation in three dimensions: structural interpretation, prior rationality, and evidence stability.

[0083] After obtaining the consistency verification record, the central coordinating agent explicitly models the conflict relationships between candidate phases by constructing a conflict graph structure, using candidate phases as nodes and conflict relationships as edges. Conflict relationships include competition around the same set of key spectral peaks, representing situations where multiple candidates interpret the same set of key spectral peaks but with different interpretation qualities; mutual exclusion between different sets of key spectral peaks, representing situations where the key spectral peak sets that each candidate phase depends on are structurally difficult to simultaneously hold true; and incompatibility between candidate phases and prior information about the sample, representing situations where candidate phases conflict with prior information about the sample at the physical, chemical, or experimental level. The conflict graph structure quantifies the conflict intensity by introducing weighted edges between nodes, where the edge weights are jointly determined by the peak coverage overlap, the difference in matching metrics, and the strength of prior violation, transforming abstract conflict relationships into computable arbitration inputs.

[0084] When performing conflict arbitration based on the conflict graph structure, the central coordinating agent ranks candidate phases in the order of prior decision, spectral peak decision, and spectral library decision. In the prior decision stage, the prior consistency verification result serves as a hard or strong constraint, directly eliminating or significantly downgrading candidate phases that do not conform to the prior information of the samples, ensuring that the final result does not violate known objective conditions. In the spectral peak decision stage, the spectral peak consistency verification results are compared among the remaining candidate phases, prioritizing those that can fully explain the key spectral peak set and have a reasonable residual distribution, thereby strengthening the dominant role of structural evidence. In the spectral library decision stage, the spectral library consistency verification results are further utilized to suppress candidate phases with unstable cross-spectral library support or high dependence on low-confidence spectral libraries, ensuring that the final ranking result remains stable and reliable at the multi-source evidence level. Through staged adjudication, different types of conflicts are resolved sequentially, forming a clear arbitration ranking result for candidate phases.

[0085] When performing backtracking verification based on the arbitration ranking results, the central coordinating agent selects the top preset number of candidate phases from the ranking results and remaps the corresponding candidate templates to the spectral axis index mapping relationship of the preprocessed spectral data set to restore the direct interpretation relationship between the candidate phases and the actual observed spectra. During backtracking verification, the focus is on verifying whether the interpretation residuals at the key spectral peak set meet the preset confidence conditions. These preset confidence conditions limit the residual amplitude, residual continuity, and residual distribution pattern within the neighborhood of the key spectral peaks, preventing situations that were ignored during the arbitration stage but show significant mismatch in the original spectral space. When a candidate phase fails the confidence conditions during backtracking verification, it is marked as a backtracking failure and removed or downgraded from the final candidates, thus forming a closed-loop verification consistent with the arbitration and data space.

[0086] After successful backtracking verification, the central coordinating agent performs structured encapsulation processing on the arbitrated phase inference results, generating a final set of analysis results. During structured encapsulation, the target phase identifier, confidence level, supporting peak set, key peak set, candidate template identifier set, consistency verification record summary, and conflict arbitration path information are uniformly written into the result entries. The spectral axis index mapping relationship and structural position mapping relationship between the final analysis result set, the peak parameter set, the preprocessed spectral data set, and the original spectral data set remain unchanged. This ensures that the output structured analysis results can be directly used for application-layer decision-making and support complete evidence backtracking and result verification for any conclusion.

[0087] This embodiment also discloses a multi-agent-based autonomous spectral data analysis device, referring to... Figure 2 The device includes an acquisition module 201, a processing module 202, and an output module 203. It is used to execute any of the above-described multi-agent-based autonomous spectral data analysis methods, wherein: The acquisition module 201 is used to acquire the original spectral data corresponding to the target analysis object and bind a preset identifier to form a set of original spectral data. Processing module 202 is used to control the quality diagnosis intelligent agent to perform data quality assessment processing on the original spectral data set and generate data quality assessment results corresponding to each original spectral data. The processing module 202 is used to control the central coordinating agent to decompose the analysis task based on the data quality assessment results, and determine the preprocessing constraints and processing intensity boundaries that match the data quality assessment results, thereby generating a set of preprocessing strategies corresponding to the quality-labeled spectral data set. The quality-labeled spectral data set controls the binding of the data quality assessment results with the original spectral data set. Processing module 202 is used to perform spectral preprocessing on the original spectral data set according to the preprocessing strategy set to obtain a preprocessed spectral data set, wherein the structural relationship between the preprocessed spectral data set and the original spectral data set is maintained during the preprocessing process; Processing module 202 is used to control the peak fitting agent to perform peak analysis processing on each preprocessed spectral data based on the preprocessed spectral data set, and generate a set of spectral peak parameters corresponding to each preprocessed spectral data. The processing module 202 is used to control the phase identification agent to match the set of spectral peak parameters with the template information in the multi-source reference spectral library, and combine the sample prior information and the confidence weight of the spectral library to generate a set of phase inference results corresponding to each candidate. The output module 203 is used to control the central coordinating agent to perform consistency verification and conflict arbitration on the set of spectral peak parameters and the set of phase inference results, form the final set of analysis results, and output them as structured analysis results.

[0088] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0089] This embodiment also discloses an electronic device, as shown in the reference. Figure 3 The electronic device may include: at least one processor 301, at least one communication bus 302, user interface 303, network interface 304, and at least one memory 305.

[0090] The communication bus 302 is used to enable communication between these components.

[0091] The user interface 303 may include a display screen and a camera. Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0092] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0093] The processor 301 may include one or more processing cores. The processor 301 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 305, and by calling data stored in memory 305. Optionally, the processor 301 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 301 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 301 and may be implemented as a separate chip.

[0094] The memory 305 may include random access memory (RAM) or read-only memory. Optionally, the memory may include a non-transitory computer-readable storage medium. The memory 305 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 305 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch functionality, sound playback functionality, image playback functionality, etc.), instructions for implementing the various method embodiments described above, etc.; the data storage area may store data involved in the various method embodiments described above, etc. Optionally, the memory 305 may also be at least one storage device located remotely from the aforementioned processor 301. As a computer storage medium, the memory 305 may include an operating system, a network communication module, a user interface 303 module, and an application program for a multi-agent-based autonomous spectral data analysis method.

[0095] exist Figure 3In the electronic device shown, the user interface 303 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 301 can be used to call an application program stored in the memory 305 that is a multi-agent-based autonomous analysis method for spectral data. When executed by one or more processors 301, the electronic device executes one or more methods as described in the above embodiments.

[0096] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps can be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0097] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0098] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.

[0099] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0100] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0101] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory 305 and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned memory 305 includes various media capable of storing program code, such as a USB flash drive, external hard drive, magnetic disk, or optical disk.

[0102] The present invention also discloses a non-transitory computer-readable storage medium storing instructions. When executed by one or more processors 301, these instructions cause an electronic device to perform one or more methods as described in the above embodiments.

[0103] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and the disclosure of practical truths. This invention is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A multi-agent-based autonomous analysis method for spectral data, characterized in that, The method includes: Obtain the raw spectral data corresponding to the target analysis object and bind it with a preset identifier to form a raw spectral data set; The quality diagnostic intelligent agent performs data quality assessment processing on the original spectral data set to generate data quality assessment results corresponding to each original spectral data. The central coordinating agent decomposes the analysis task based on the data quality assessment results and determines the preprocessing constraints and processing intensity boundaries that match the data quality assessment results, thereby generating a set of preprocessing strategies corresponding to the quality-labeled spectral data set. The quality-labeled spectral data set is formed by binding the data quality assessment results with the original spectral data set. Based on the set of preprocessing strategies, spectral preprocessing is performed on the original spectral data set to obtain a preprocessed spectral data set, wherein the structural relationship between the preprocessed spectral data set and the original spectral data set is maintained during the preprocessing process; Based on the preprocessed spectral data set, the peak fitting agent performs peak analysis processing on each preprocessed spectral data to generate a set of peak parameters corresponding to each preprocessed spectral data. The phase identification agent matches the set of spectral peak parameters with the template information in the multi-source reference spectral library, and combines the sample prior information with the confidence weight of the spectral library to generate a set of phase inference results corresponding to each candidate. The central coordinating agent performs consistency verification and conflict arbitration on the set of spectral peak parameters and the set of phase inference results to form the final set of analysis results, which is then output as structured analysis results.

2. The autonomous spectral data analysis method based on multi-agent systems according to claim 1, characterized in that, The process of performing data quality assessment on the original spectral data set by the quality diagnostic intelligent agent to generate data quality assessment results corresponding to each original spectral data set specifically includes: A quality assessment context is established for each of the original spectral data, and the quality assessment context includes at least an acquisition condition identifier, an instrument status identifier, and a time identifier; Under the constraints of the quality assessment context, the original spectral data are jointly modeled based on the global amplitude distribution, local fluctuation distribution, and noise ratio of the spectral intensity sequence to extract signal-to-noise features that characterize the level of random noise. By fitting the slowly varying trend term within a preset spectral axis range to each original spectral data, and evaluating the continuity and consistency of the fitting residual as the spectral axis position changes, a baseline stability feature is generated to characterize the degree of baseline drift. By analyzing the intensity abrupt change relationship between sampling points on adjacent spectral axes and the probability of repeated occurrence of abrupt change points within a local window, spike noise features used to characterize isolated spikes and outlier perturbations are extracted from the original spectral data. The original spectral data are processed by detecting whether the spectral intensity reaches the upper limit of the instrument range within the continuous spectral axis range, and spectral saturation features are generated to characterize the risk of effective information truncation. Based on the alignment relationship between multiple original spectral data in the original spectral data set, the relative position deviation of characteristic peaks in different original spectral data is analyzed to generate spectral axis consistency features to characterize the stability of the spectral axis. The signal-to-noise ratio feature, baseline stability feature, spike noise feature, spectral line saturation feature, and spectral axis consistency feature are mapped to a unified evaluation scale and assigned corresponding weight parameters according to a preset quality evaluation strategy to generate data quality evaluation results corresponding to each of the original spectral data.

3. The autonomous spectral data analysis method based on multi-agent systems according to claim 1, characterized in that, The central coordinating agent decomposes the analysis task based on the data quality assessment results and determines preprocessing constraints and processing intensity boundaries that match the data quality assessment results, thereby generating a set of preprocessing strategies corresponding to the quality-labeled spectral data set, specifically including: The data quality assessment results are parsed to identify the corresponding quality feature type, anomaly degree, and repair level in each data quality assessment result, thereby generating a corresponding quality constraint description for each of the original spectral data. Based on the quality constraint description, the overall spectral analysis task is decomposed into preprocessing subtasks associated with data quality characteristics. Each of the preprocessing subtasks is bound to a corresponding quality trigger condition, so that each of the preprocessing subtasks is activated when the corresponding quality anomaly condition is met; Based on the quality constraint description, preprocessing constraints are determined for each preprocessing subtask. These preprocessing constraints are used to define the boundaries by which the preprocessing subtask affects the spectral structure during execution. Based on the quantification level of each quality feature in the data quality assessment results, a processing intensity boundary is determined for each preprocessing subtask. The processing intensity boundary is used to limit the range of parameter adjustment in the corresponding preprocessing subtask. The preprocessing subtasks, preprocessing constraints, and processing intensity boundaries are jointly scheduled and arranged according to the data dependencies and structural influence order among the preprocessing subtasks, and a preprocessing decision scheme is generated. Based on the preprocessing decision scheme, the preprocessing subtasks, the preprocessing constraints, and the processing intensity boundaries are combined to generate a set of preprocessing strategies corresponding to the quality-labeled spectral data set.

4. The autonomous spectral data analysis method based on multi-agent systems according to claim 1, characterized in that, The step of performing spectral preprocessing on the original spectral data set according to the preprocessing strategy set to obtain a preprocessed spectral data set specifically includes: A preprocessing execution context is established based on the preprocessing strategy set, and the preprocessing execution context is used to limit the execution order of preprocessing subtasks; Under the constraints of the preprocessing execution context, spectral preprocessing operations are performed sequentially for each of the original spectral data in the execution order. When performing the baseline correction preprocessing subtask, the slowly varying trend term is modeled and corrected according to the preprocessing constraints. When performing the noise suppression preprocessing subtask, random noise components are suppressed in a restricted manner according to the processing intensity boundary; When performing the peak repair preprocessing subtask, local replacement or reconstruction processing is performed on abnormal sampling points; When performing the spectral axis alignment adjustment preprocessing subtask, normalization is performed without changing the number and arrangement order of spectral axis sampling points; After completing each preprocessing subtask, the processed spectral data are summarized in a predetermined order to generate the preprocessed spectral data set.

5. The autonomous spectral data analysis method based on multi-agent systems according to claim 1, characterized in that, Based on the preprocessed spectral data set, the peak fitting agent performs peak analysis processing on each preprocessed spectral data to generate a set of peak parameters corresponding to each preprocessed spectral data set, specifically including: To establish a spectral peak analysis context based on the preprocessed spectral data set, the spectral peak analysis context shall at least include spectral axis index mapping relationships, key structural position mapping relationships, and preprocessing process records; Under the constraints of the spectral peak analysis context, initial spectral peak detection processing is performed on each preprocessed spectral data. By jointly analyzing the local extreme value features, gradient change features, and curvature change features of the spectral intensity along the spectral axis, a set of potential spectral peak candidate positions is generated. Based on the spectral morphology information in the peak analysis context, an adaptive peak number determination process is performed on each preprocessed spectral data. By analyzing the spacing relationship, overlap degree, and local signal-to-noise level between the candidate peak positions in the candidate peak position set, the target peak number corresponding to each preprocessed spectral data is determined. Initial spectral peak parameters are generated for each target spectral peak. These initial spectral peak parameters include peak position parameters, peak width parameters, peak shape parameters, and peak intensity parameters. The value range of the initial spectral peak parameters is jointly limited by the structure preservation constraints in the spectral peak analysis context and the preprocessing process records. The initial spectral parameters are iteratively adjusted by constructing an objective function that includes a spectral peak superposition model and a background term for each of the preprocessed spectral data, and a robust loss constraint is introduced during the fitting process to output the fitting result. The fitting results are compared with the preprocessed spectral data in terms of overall contour, local residual distribution and key structural positions. When the spectral peak analysis results meet the preset reliability conditions, the final spectral peak parameters are output. A set of spectral peak parameters corresponding to each of the preprocessed spectral data is generated for the final spectral peak parameters.

6. The autonomous spectral data analysis method based on multi-agent systems according to claim 1, characterized in that, The phase identification agent matches the set of spectral peak parameters with template information in a multi-source reference spectral library, and combines prior sample information with library confidence weights to generate a set of phase inference results corresponding to each candidate, specifically including: The multi-source reference spectrum library is used to select a set of candidate templates that are comparable to the set of spectral peak parameters in terms of structural features based on the distribution range of peak position parameters, peak number characteristics and spectral band coverage relationship. By establishing the correspondence between the peak position parameters in the spectral peak parameter set and the characteristic peak positions of the candidate templates in the candidate template set, a peak position alignment mapping relationship is formed; Based on the peak alignment mapping relationship, the matching measurement results are generated for each candidate template by jointly calculating the peak position deviation, peak width matching degree, peak shape similarity, and peak intensity ratio consistency. The matching metric results are constrained and corrected by introducing prior information of the sample. A first template is obtained by suppressing or eliminating candidate templates that do not conform to the prior information of the sample. The first template is weighted and fused by introducing spectral library confidence weights to obtain the second template; A comprehensive inference process is performed on the candidate phases corresponding to the second template to generate phase inference results corresponding to each candidate phase, forming the set of phase inference results.

7. The autonomous spectral data analysis method based on multi-agent systems according to claim 1, characterized in that, The central coordinating agent performs consistency verification and conflict arbitration on the set of spectral peak parameters and the set of phase inference results to form a final set of analysis results, which is then output as structured analysis results, specifically including: Construct corresponding arbitration contexts for each of the candidate phases; Under the arbitration context constraints, the spectral peak parameter set is subjected to spectral peak evidence reconstruction processing and mapped to a spectral peak evidence set; A consistency verification process is performed on the set of phase inference results to generate a consistency verification record corresponding to each candidate. The consistency verification process includes peak consistency verification for key peak sets, prior consistency verification for sample prior information, and library consistency verification for the stability of multi-source reference spectral libraries. Based on the consistency verification record, the conflict relationship between candidate phases is explicitly modeled. By identifying the competitive relationship between candidate phases around the same key spectral peak set, the mutual exclusion relationship between different key spectral peak sets, and the incompatibility relationship between candidate phases and sample prior information, a conflict graph structure is constructed. Based on the conflict graph structure, conflict arbitration processing is performed to form an arbitration ranking result of candidate phases. The conflict arbitration processing ranks the candidate phases in the order of prior decision, spectral peak decision, and spectral library decision. Based on the arbitration ranking results, a backtracking verification process is performed on the first preset number of candidate phases. This involves remapping the corresponding candidate templates to the spectral axis index mapping relationship of the preprocessed spectral data set and verifying whether the interpretation residuals at the key spectral peak sets meet the preset confidence conditions. After the backtracking verification is passed, the phase inference results that have passed the arbitration are encapsulated in a structured manner to form the final set of analysis results.

8. A multi-agent-based autonomous spectral data analysis device, characterized in that, The device is used to execute a multi-agent-based autonomous spectral data analysis method as described in any one of claims 1-7, the device comprising an acquisition module, a processing module, and an output module, wherein: The acquisition module is used to acquire the original spectral data corresponding to the target analysis object and bind a preset identifier to form a set of original spectral data. The processing module is used to control the quality diagnostic agent to perform data quality assessment processing on the original spectral data set and generate data quality assessment results corresponding to each original spectral data. The processing module is used to control the central coordinating agent to decompose the analysis task according to the data quality assessment result, and determine the preprocessing constraints and processing intensity boundaries that match the data quality assessment result, thereby generating a set of preprocessing strategies corresponding to the quality-labeled spectral data set, wherein the quality-labeled spectral data set is formed by binding the data quality assessment result with the original spectral data set; The processing module is used to perform spectral preprocessing on the original spectral data set according to the preprocessing strategy set to obtain a preprocessed spectral data set, wherein the structural correlation between the preprocessed spectral data set and the original spectral data set is maintained during the preprocessing process; The processing module is used to control the peak fitting agent to perform peak analysis processing on each preprocessed spectral data based on the preprocessed spectral data set, and generate a set of spectral peak parameters corresponding to each preprocessed spectral data. The processing module is used to control the phase identification agent to match the set of spectral peak parameters with the template information in the multi-source reference spectral library, and combine the sample prior information and the confidence weight of the spectral library to generate a set of phase inference results corresponding to each candidate. The output module is used to control the central coordinating agent to perform consistency verification and conflict arbitration on the set of spectral peak parameters and the set of phase inference results, form the final set of analysis results, and output them as structured analysis results.

9. An electronic device, characterized in that, The device includes a processor, a communication bus, a user interface, a network interface, and a memory. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The communication bus is used to enable communication between the components within the electronic device. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1-7.

Images