Multi-dimensional operation analysis index automatic adjustment and optimization management method

Abstract

The invention relates to the technical field of operation data analysis, in particular to a multi-dimensional operation analysis index automatic adjustment and optimization management method, which comprises the following steps of: connecting a plurality of service systems through a distributed data acquisition gateway, and constructing a standardized data pool based on a dynamic pattern recognition algorithm; calling a dynamic index calculation engine, performing multi-dimensional feature extraction on the standardized data pool according to a preset index generation rule, and generating an initial index set; inputting the initial index set into an optimization model, generating an index optimization scheme through an association constraint analysis module, and outputting an optimization index set; and establishing an index dynamic update link. According to the invention, the health state of the system is monitored in real time in the release process of each business index, rollback control is triggered based on the change trend of the health index, and the updating safety is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of business data analysis, and particularly to a method for automatically adjusting and optimizing multi-dimensional business analysis indicators and management methods. Background Art

[0002] With the continuous improvement of the informatization level of enterprises, a large amount of heterogeneous data resources have been accumulated in various business systems (such as financial systems, supply chain systems, customer relationship management systems, etc.). Traditional business analysis relies on fixed report templates or manually formulated indicator systems, often lacking the ability to respond to dynamic business changes, resulting in lagging indicator calibers, accumulation of redundant indicators, frequent analysis misjudgments, and difficulty in supporting the needs of refined business decisions.

[0003] In the prior art, some business systems introduce data middle platforms or BI tools for indicator management, but generally have the following problems: First, the cross-system data integration ability is weak, and it is difficult to efficiently integrate in the face of situations such as inconsistent field naming and data interface protocols; second, the indicator system lacks context awareness ability and cannot be automatically adjusted as the business scenario evolves; third, the indicator optimization process highly depends on manually set rules and lacks a structured modeling and constraint consistency verification mechanism; fourth, the indicator update process lacks version control and gray release capabilities, which is prone to risks of system interference and business shocks. Summary of the Invention

[0004] The present invention provides a method for automatically adjusting and optimizing multi-dimensional business analysis indicators, a management method that is oriented to multiple business systems and has the ability to automatically generate and adaptively optimize indicators, constructs a unified data expression and indicator modeling mechanism, and integrates key technologies such as graph structure analysis, context awareness, and version collaboration to achieve the automatic adjustment and optimization management of multi-dimensional business analysis indicators.

[0005] The method for automatically adjusting and optimizing multi-dimensional business analysis indicators includes the following steps: S1: Connect to multiple business systems through a distributed data collection gateway, and construct a standardized data pool based on a dynamic pattern recognition algorithm; S2: Invoke a dynamic indicator calculation engine, perform multi-dimensional feature extraction on the standardized data pool according to preset indicator generation rules, and generate an initial indicator set; S3: Input the initial indicator set into an optimization model, generate an indicator optimization plan through an association constraint analysis module, and output an optimized indicator set; S4: Distribute the optimized indicator set to each business system through a version collaboration controller, and establish a dynamic indicator update link.

[0006] Optionally, the S1 specifically includes: Decouple the features of the data interfaces of multi-service systems through a heterogeneous protocol parser to generate a unified data schema template; Adopt a dynamic sliding window mechanism to capture the time-varying features of the data stream, and identify the mutation points of the data pattern based on the LSTM-Attention model; Construct a data fingerprint library including spatio-temporal two-dimensional features, and eliminate the semantic differences across business systems through a fuzzy matching algorithm; Based on the pattern recognition results, trigger an adaptive cleaning rule engine to generate a standardized data pool with version traceability identifiers, where the data items include: Metadata blood relationship graph: Identify the original data source and conversion path, generated in the protocol feature decoupling stage, and formed by recording the field source and parsing path; Dynamic confidence score: Reflect the data quality level, calculated by the confidence score formula in the standardized data pool construction stage; Cross-business system association anchor: Used for index association analysis, extracted through field matching relationships in the data fingerprint fuzzy matching stage.

[0007] Optionally, the S2 specifically includes: S21, construct an index generation rule parser, convert the preset index rule set into a unified syntax tree structure, each rule includes an operator sequence, a data source mapping, and an execution condition predicate after parsing, and establish an executable and interpretable calculation flow structure to support rule triggering and index derivation; S22, perform multi-modal feature extraction processing on different types of data in the standardized data pool, and fuse each modal feature to form a unified index feature expression space; S23, construct an association graph between indexes based on the index feature expression space, calculate the overall association strength between indexes using a multi-source heterogeneous similarity model, fuse the feature space similarity, the time series trajectory similarity, and the data source structure similarity, and introduce a structure importance index for pruning for the generated association edge set, only retain the edge relationships corresponding to high similarity or key structural paths to form a sparse optimized index graph structure; S24, output an initial index set that meets the validity conditions, and retain its structural adjacency information in the association graph, and this set is used as the input basis for subsequent index management, visualization display, or adaptive regulation mechanism.

[0008] Optionally, the multi-modal feature extraction processing in the S22 specifically includes: For numerical data, extract key features by combining statistical dimensionality reduction and signal transformation methods; For text data, obtain semantic expression vectors through deep semantic modeling and keyword weight combination; For time-series data, construct a feature channel based on time-series dependency modeling to obtain the dynamic evolution features of the context sequence.

[0009] Optionally, S2 further includes combining scene context information to perform dynamic weight assignment on the metric set, calculating the context importance score of each metric through the similarity between the feature vector and the scene vector, executing the metric validity verification process, and screening out unrepresentative or redundant metrics based on the statistical deviation and information entropy of the feature residuals to obtain valid metrics.

[0010] Optionally, the initial metric set is represented as: , where represents the initial metric set, is a valid metric, represents the set of associated metrics adjacent to metric in the metric map, represents the context-aware weight value of metric , which is used to measure the importance or priority of this metric under the current scene features, represents the set of all valid metrics determined to be valid, which is the inverse mapping of the metric indices for which the result of the validity determination function is 1.

[0011] Optionally, S3 specifically includes: S31, construct a metric constraint relation graph, form an enhanced adjacency matrix representing the logical constraint relations between metrics, convert it into a constraint strength using a multi-layer perceptron network by parsing the rule description vector, and combine the basic direction relations between metric pairs to form a weighted adjacency matrix to characterize the directed constraint dependencies between each metric; S32, based on the enhanced adjacency matrix, perform conflict path detection operations, use the path synthesis algorithm to analyze whether there are path conflicts between any two metrics, and identify potential path connections that violate logical constraints by introducing a direction consistency discrimination mechanism; S33, after the conflict path detection operation, construct a multi-objective optimization function including the main objective and constraint violation terms. The multi-objective optimization function takes the deviation between the metric function output and the target value as the main term, and introduces penalty terms for all identified valid constraints. The constraint penalty strength is dynamically adjusted according to the violation frequency and the optimization process to form an adaptive regulation mechanism; S34, perform gradient optimization on the weight variables of each metric based on the metric constraint relation graph. The gradient calculation considers the adjacency relationship of the metric in the graph and is propagated through the influence factors accumulated along the path, retaining the structural constraint dependencies in the weight update process to ensure topological consistency in the adjustment process.

[0012] Optionally, S3 further includes performing double verification on the stability and consistency of all updated metrics, determining whether the weight change is within the set range, and simultaneously evaluating the logical consistency level under all associated constraints. When both conditions are met, the current metric is incorporated into the optimized metric set and the optimized metric set is output. Metrics that fail the verification will be marked as requiring recalculation status.

[0013] Optionally, S4 specifically includes: S41, generating an incremental update package: performing a difference analysis on the old and new metric sets, calculating the change values of the weights of each metric, forming a difference set consisting of metric identifiers and weight differences, and performing difference-driven coding compression on the difference set. Specifically, it includes representing the metric identifiers in a hashed form, applying ZigZag coding to the weight differences, connecting the results after coding, and appending an integrity check code to form an incremental update package. By calculating the ratio of the total length after coding to the length of the original metric set, the compression rate is calculated for subsequent evaluation of the release conditions; S42, executing an improved transaction coordination protocol: The transaction update process includes three stages: pre-commit, compensation negotiation, and final commit; S43, implementing a gray release strategy: Based on the current load level and criticality of each business unit, calculate its update priority, and complete batch division based on this priority. The number of business units included in each batch is related to the current system load pressure, and the total number of batches is adjusted to achieve load balancing. During the batch update process, the metric update tasks are deployed batch by batch in descending order of priority; S44, constructing a health-driven rollback mechanism: During the gray release period, continuously monitor multiple key performance indicators, and calculate the overall health index of the business system based on their current values and normal reference values. The overall health index combines each sub-index through geometric weighting to reflect the current operating state of the business system. When it is detected that the overall health index continues to decline, the change rate exceeds the warning threshold, and at the same time is lower than the minimum health tolerance value, it is determined to enter the degradation state, and the rollback operation of the metric version is automatically triggered to ensure business continuity and data consistency.

[0014] Optionally, in the pre-commit stage, a request message including the incremental update package and timestamp is constructed and distributed to all business systems to publish pre-update information (based on the compression rate evaluation). After each business subsystem receives the request, it returns a signature confirmation response. After the coordinator collects multiple valid responses, it determines that the legal response condition is met. If the response is insufficient within the preset time, it enters the compensation stage and executes an exponential backoff retry strategy based on the time consumption and the number of failures. After the compensation attempt fails, the update process is terminated. If the response meets the conditions, the final commit stage is executed, and a final transaction message including signature information and incremental data is constructed. Each business system performs a consistency check after receiving it to ensure that the updated data content is exactly the same as the source data.

[0015] Advantages of the present invention: In the present invention, a heterogeneous protocol parser is used to connect to the data interfaces of multiple business systems, automatically complete field semantic mapping and unified data schema construction, combine the LSTM-Attention model to perform dynamic pattern recognition and mutation detection on data streams, and introduce a data fingerprint fuzzy matching algorithm to eliminate cross-system semantic ambiguity. Finally, a standardized data pool containing metadata lineage graphs, confidence scores, and associated anchors is constructed. Compared with the traditional method that relies on manually defining field mapping rules, it has higher adaptability and generalization ability, effectively solves problems such as inconsistent multi-system data formats, chaotic field naming, and difficult cross-domain matching, and provides a high-quality data basis for subsequent indicator mining and optimization.

[0016] In the present invention, an index generation rule parser is constructed to realize a traceable and executable index generation process; through multi-modal feature fusion such as numerical, text, and time series types, a unified index feature expression space is constructed, and a hybrid similarity modeling, graph pruning, and context-aware weight allocation mechanism are introduced to form an initial index set with sparse structure, clear semantics, and adaptable to scenario changes. In the optimization stage, an enhanced adjacency matrix is used to construct a constraint graph, and a multi-objective function and a graph structure gradient propagation mechanism are fused to realize dynamic adjustment of index weights under constraint consistency control. Compared with the existing management method that relies on static index templates, the present invention can adaptively optimize index combinations according to business status, significantly improving the accuracy and controllability of the analysis system.

[0017] In the present invention, an index update mechanism driven by incremental compression is constructed, and lightweight index publishing is realized through differential encoding and compression rate judgment; an improved three-phase transaction coordination protocol is used to ensure the consistency and fault tolerance of distributed publishing, and gray-scale distribution control is realized by combining dynamic batch scheduling and priority scoring. During the index publishing process, the system health status is monitored in real time, and rollback control is triggered based on the change trend of the health index to improve update security. Brief Description of the Drawings

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 It is a schematic flowchart of the method according to an embodiment of the present invention; Figure 2 It is a schematic diagram of the generation of the initial index set according to an embodiment of the present invention. Detailed Embodiments

[0020] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; moreover, the accompanying drawings are only for more specifically describing the embodiments, and are not intended to specifically limit the present invention.

[0021] It should be noted that in the specification, when referring to "an embodiment", "embodiment", "exemplary embodiment", "some embodiments", etc., it indicates that the described embodiment may include specific features, structures or characteristics, but not necessarily every embodiment includes such specific features, structures or characteristics. Additionally, when combining embodiments to describe specific features, structures or characteristics, implementing such features, structures or characteristics in combination with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the relevant art.

[0022] Generally, terms can be understood at least in part from their use in context. For example, at least in part depending on the context, the term "one or more" used herein can be used to describe any feature, structure or characteristic in a singular sense, or can be used to describe a combination of features, structures or characteristics in a plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey a set of exclusive factors, but rather, at least in part depending on the context, can allow for the existence of other factors that are not necessarily explicitly described.

[0023] As Figure 1 - Figure 2 shown, the multi-dimensional operation analysis index automatic adjustment and optimization management method includes the following steps: S1: Connect to multiple business systems through a distributed data acquisition gateway, and build a standardized data pool based on a dynamic pattern recognition algorithm; S2: Invoke a dynamic index calculation engine, perform multi-dimensional feature extraction on the standardized data pool according to preset index generation rules, and generate an initial index set; S3: Input the initial index set into an optimization model, generate an index optimization plan through an association constraint analysis module, and output an optimized index set; S4: Distribute the optimized index set to each business system through a version collaboration controller, and establish an index dynamic update link.

[0024] S1 specifically includes: Perform feature decoupling on the data interfaces of multiple business systems through a heterogeneous protocol parser to generate a unified data mode template; Protocol feature decoupling: Perform feature decoupling on the data interfaces of multiple business systems through a heterogeneous protocol parser to generate a unified data mode template; the protocol parser processes the original data field set of the Perform the following operations: , where Embedding(·) is a word vector mapping constructed based on the field naming pattern, represents the vector concatenation operation, and LSTM(·) represents a long short-term memory network extractor used to extract the semantic features of the field sequence and generate a protocol feature vector , is the th field in the th system, represents the th original field set of the Capture the time-varying features of the data stream using a dynamic sliding window mechanism, and identify the mutation points of the data pattern based on the LSTM-Attention model; Time-varying feature capture: Analyze the time characteristics of the data stream using a dynamic sliding window mechanism. The window length is adaptively adjusted according to the data volatility within the window. The calculation formula is: , where is the basic window length, is the standard deviation of the data within the current window, is the mean of the data within the current window.

[0025] Identification of data pattern mutations: Identify the mutation points of the data stream based on the LSTM-Attention model, specifically including: Calculation of attention weights: ; Update of the LSTM cell state: ; ; ; ; ; ; Among them, represents the current input feature, represents the current LSTM hidden state, represents the previous LSTM hidden state, represents the current moment attention weight, , , represents the Attention weight matrix, , , Denote the LSTM input gate, forget gate, and output gate, Denote the candidate memory cell, Denote the memory cell state, , , , Denote the bias term, Denote the Sigmoid activation function, Denote the hyperbolic tangent function, Denote the Hadamard element-wise multiplication; Pattern mutation point trigger condition: ; Among them, Is the Euclidean distance of the LSTM hidden state, Denote the hidden state difference threshold, Denote the attention weight threshold; Construct a data fingerprint library including spatio-temporal two-dimensional features, and eliminate semantic differences across business systems through a fuzzy matching algorithm; Data fingerprint construction and fuzzy matching: Construct a data fingerprint library containing spatio-temporal two-dimensional features, and the fingerprint of the field Is represented as: ; Among them, Is the time series feature of the field, Is the spatial association feature of the field; Denote the vector concatenation operation, Denote the average value, Denote the feature concatenation, FFT() is the fast Fourier transform, Denote the field The fingerprint value of, Hash() is the feature compression hash operation, and the Jaccard similarity is used for fuzzy matching to calculate the similarity of any field pair : : , where co-occur(A,B) is the co-occurrence times of fields A and B, and N is the total number of data batches; Based on the pattern recognition results, trigger the adaptive cleaning rule engine to generate a standardized data pool with version traceability identifiers, where the data items include: Metadata lineage graph: Identify the original data source and conversion path, generated in the protocol feature decoupling stage, and formed by recording the field source and parsing path; Dynamic confidence score: Reflect the data quality level, calculated by the confidence score formula in the standardized data pool construction stage; Cross-business system association anchor: Used for indicator association analysis, it is obtained by extracting through field matching relationships during the data fingerprint fuzzy matching stage.

[0026] Standardized data pool construction and confidence scoring: Confidence scores Confidence are assigned to each piece of data in the standardized data pool. The scoring formula is as follows: + + , where is the data update time interval, is the data missing rate (integrity), is the number of missing data items, Total is the total number of data items, is the cross-system consistency ratio, is the number of items passing consistency, CrossCheck represents the number of systems participating in the consistency check, represents the freshness decay factor, , , is the confidence weight coefficient, which is adaptively updated through online learning: , where is the cross-entropy loss function constructed based on the data quality verification results, is the learning rate, represents the th iteration of the th weight. Finally, a standardized data pool containing elements such as metadata lineage graphs, dynamic confidence scores, and cross-business system association anchors is generated, and version tracing identifiers are bound to support subsequent indicator evolution management.

[0027] S2 specifically includes: S21, constructing an indicator generation rule parser: Assume the rule library contains rules: ; Each rule can be parsed into a syntax tree structure: , where represents the operator sequence (operator chain), is the data source mapping relied on by the rule, is the predicate logic composed of the rule execution conditions, represents the th indicator generation rule, which is a specific rule in the rule library, represents the syntax tree structure parsed by rule , including three components: the operator sequence, data source mapping, and execution condition predicate of the rule.

[0028] S22. Perform joint extraction of multi-dimensional features. For different modal data in the standardized data pool, extract multiple types of features: Numeric features Extract: ; Text features Extract: ; Temporal features Extract: ; Among them, represents the feature concatenation operation, represents the feature cross operation, is the standardized numeric data, is the original signal data, is the text data, represents the temporal window data.

[0029] S23. Construct an index association graph: Calculate the association strength between indexes through a hybrid similarity model: ; Among them, represents the index and the index The comprehensive association strength between them, represents the cosine similarity of the feature vector, represents the dynamic time warping distance of the index temporal trajectory, is the Jaccard similarity of the index data source set, , is the similarity weighting coefficient, which supports dynamic setting according to the scenario, and takes values 0.5, 0.3, and 0.2 respectively; Graph dynamic pruning strategy: , among them, is the similarity retention threshold, taking the value 0.75, is the edge betweenness centrality index, is the structural importance threshold, taking values from 0.05 to 0.1, represents the set of edges retained in the index association graph, including all index pairs that meet the similarity threshold or structural importance conditions.

[0030] S24. Perform dynamic weight assignment and validity verification: Index weight calculation (based on context feature vector): , among them, represents the index feature vector, Represents the context feature vector in the current scenario, which is the normalized index weight.

[0031] Index validity verification: ; Among them, represents the current index feature, is the historical fitting prediction value, is the standard deviation of the index feature fluctuation, is the residual tolerance threshold, represents the index information entropy, is the lower limit of effective performance, with a value of 0.5, indicating the index validity determination function. If both the residual deviation limit and the information entropy lower limit conditions are met, it returns 1 (valid), otherwise it returns 0 (invalid).

[0032] S25, output the initial index set with topological relationship, expressed as (that is, construct the effective indexes into a structured set and output): , among which, represents the initial index set, is the effective index, represents the set of associated indexes adjacent to the index in the index map, represents the context-aware weight value of the index , used to measure the importance or priority of this index under the current scenario features, represents the set of all effective indexes determined to be valid, which is the inverse mapping of the index indexes with the result of 1 in the validity determination function .

[0033] S3 specifically includes: S31, construct the index constraint relationship graph: Represent the constraint relationship between indexes by enhancing the adjacency matrix: , among which, represents the basic constraint relationship direction vector from the index to , represents the rule description vector (text or structured constraint expression) between them, represents the constraint strength function, defined as: , where MLP is a multi-layer perceptron, is the activation function, is the trainable weight vector.

[0034] S32, Conflict Path Detection: Based on the enhanced adjacency matrix, use the Floyd-Warshall (path synthesis algorithm) algorithm to calculate the potential conflict paths between metrics: , where are two metrics for which it is to be detected whether there are conflict paths, is the enhanced adjacency matrix representing the metric constraint relationship, represents the metric to constraint strength, represents the metric to constraint strength, represents taking the union of the synthesis results for all intermediate nodes , represents the total number of metrics, is the constraint relationship synthesis operator (used to judge directional conflicts), defined as: , is the conflict detection sensitivity threshold, and Sign() represents the sign function, used to judge the consistency of the constraint direction.

[0035] S33, Multi-Objective Optimization Function Construction: Let the weight matrix of the metric set be W, and define the following multi-objective loss function: , where represents the output of the metric function obtained based on the current weight, is the target output value, is the set of effective constraints between metrics, is the violation degree function of the constraints with respect to the weights, is the dynamic weight of the main loss and constraint penalty, where: , is the adjustment factor, is the cumulative number of constraint violations in the historical optimization process, is the current optimization round number, is the time decay coefficient.

[0036] S34, Gradient-Guided Constraint Propagation Optimization: For each metric weight , calculate its weight adjustment gradient on the constraint graph: , where is the total loss function, is the current weight of the metric , is the gradient of the total loss function with respect to , is the set of metrics adjacent to the metric in the constraint graph, Represents the combined loss term involved in the constraint relationship For the partial derivative of this constraint term with respect to Represents the influence factor from To The influence factor is defined in the form of a path integral: Represents the constraint strength of each edge in the path. This mechanism ensures the preservation of the structural dependence between metrics during the weight optimization process Represents the path set in the constraint graph from To Represents the derivative of this edge weight with respect to the weight Represents the product of the derivative results of all edges in the path Represents the integration over all paths, reflecting the influence of full-path propagation

[0037] S35, Optimization result verification and screening: The final verification form of the output metric is defined as follows: Where Represents the optimized output result of the metric Is the original metric Represents the consistency metric value of the metric In the constraint graph. The consistency metric function is defined as: Is the current metric weight of the metric Is the change in the metric weight before and after optimization Is the weight stability threshold Is the constraint consistency judgment threshold Is the constraint set related to the metric Represents re-evaluating or removing unstable metrics

[0038] S4 specifically includes: S41, Generate an incremental update package: Perform a difference analysis on the new and old metric sets, calculate the change values of each metric weight, form a difference set consisting of metric identifiers and weight differences, and perform difference-driven coding compression on the difference set. Specifically, it includes hashing the metric identifiers and applying ZigZag coding to the weight differences. After coding, connect the results and append an integrity check code to form an incremental update package. Calculate the compression rate by statistically calculating the ratio of the total length after coding to the length of the original metric set for subsequent evaluation of the release conditions; ​​​​​​​​The compression ratio is used to evaluate the lightweight degree of the metric update package, so as to determine whether it is suitable for release under the current system load: High compression ratio (close to 1): It indicates that the update package has a small volume, low transmission and processing overhead, and can enter the fast release channel preferentially to reduce the risk of system interference; Low compression ratio (close to 0): It indicates that there are many changes or low coding efficiency, which may trigger release delays, split update batches or enter the gray control mode.

[0039] Compression ratio calculation: , where CompressRatio is the compression ratio, indicating the degree of compression of the new data volume of the metric update after differential coding compression, is the compressed incremental update package, which contains the identification hash of each metric and the encoded weight difference, is the total data length (in bytes or bits) of the compressed update package, is the complete information (original format) of the kth metric, is the data length occupied by the kth metric in the original format, is the total number of metrics, indicating the number of metrics currently being compressed, is the total data length of all original metric information; this formula represents the reduction ratio of the total volume of compressed data relative to the volume of the original data, and is used to measure the compression efficiency.

[0040] S42. Execute the improved transaction coordination protocol: The transaction update process includes three stages: pre-commit, compensation negotiation, and final commit; S43. Implement the gray release strategy: Based on the current load level and criticality of each business unit, calculate its update priority, and complete batch division based on this priority. The number of business units included in each batch is related to the current system load pressure, and the total number of batches is adjusted to achieve load balancing. During the batch update process, according to the priority from high to low, deploy the metric update tasks batch by batch; The update priority of the business unit is expressed as: ; Definition of batch update: ; Total number of batches: ; Among them: is the jth business unit, is the current system load of the jth business unit, and max(Load) is the maximum load value among all business units, Critj is the business criticality score for the j-th business unit, and ∑Crit is the total sum of criticality scores for all business units. UpdatePriorityj is the updated priority score for the j-th business unit. Bq is the set of business units included in the q-th batch, where q is the current batch number (starting from 0), and Q is the total number of batches, which is dynamically calculated based on the system pressure. N is the total number of business units, and SystemStress is the current overall system pressure assessment value. SystemBaseThreshold is the system benchmark pressure threshold used to normalize the system load, and log2(·) is the logarithm function with base 2. ⌈ ⌉ represents rounding up. ⌊ ⌋ represents rounding down. S44. Build a health-driven rollback mechanism: During the gray release period, continuously monitor multiple key performance indicators, and calculate the overall health index of the business system based on their current values and normal reference values. The overall health index integrates each sub-index through geometric weighting to reflect the current operating state of the business system. When it is detected that the overall health index continues to decline, and the change rate exceeds the warning threshold and is simultaneously lower than the minimum health tolerance value, it is determined that the system has entered a degraded state, and the rollback operation of the indicator version is automatically triggered to ensure business continuity and data consistency.

[0041] Overall health index of the health-driven rollback mechanism system: ; Among them, Mi(t) represents the actual value of the i-th health monitoring indicator at the current moment, including: M1 request success rate, M2 response latency, M3 resource utilization rate, M4 data consistency. Refi is the normal reference value of the i-th indicator, Π is the product symbol, and wi is the weight of each indicator, satisfying Rollback trigger conditions: ; Among them, H(t) is the overall health index of the system at the current moment t. ΔHThreshold is the warning threshold for the decline rate of the health index. HTolerance is the minimum tolerance critical value of the health index, and Rollback is the rollback flag, with a value of 1 indicating that rollback is triggered, and 0 indicating that rollback is not triggered.

[0042] In the pre-submission stage, a request message including an incremental update package and a timestamp is constructed and distributed to all business systems to publish pre-update information (based on compression ratio evaluation). After each business subsystem receives the request, it returns a signature confirmation response. After the coordinator collects multiple valid responses and determines that the legal response conditions are met, if the responses are insufficient within the preset time, the compensation stage is entered, and an exponential backoff retry strategy is executed based on the time consumption and the number of failures. After the compensation attempt fails, the update process is terminated. If the responses meet the conditions, the final submission stage is executed, and a final transaction message including signature information and incremental data is constructed. After each business system receives it, consistency verification is performed to ensure that the updated data content is exactly the same as the source data.

[0043] This invention covers any alternatives, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To enable the public to have a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments of this invention. However, those skilled in the art can fully understand this invention even without the description of these details. Additionally, well-known methods, processes, procedures, components, and circuits, etc., are not described in detail to avoid unnecessary confusion to the essence of this invention.

[0044] The above description is only a preferred embodiment of this invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of this invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of this invention.

Claims

1. A multi-dimensional business analysis indicator automatic adjustment and optimization management method, characterized in that: The following steps are involved: S1: Connect multiple business systems through distributed data collection gateways and build a standardized data pool based on dynamic pattern recognition algorithms; S2: Call the dynamic indicator calculation engine to extract multi-dimensional features from the standardized data pool according to the preset indicator generation rules to generate an initial indicator set; S3: Input the initial indicator set into the optimization model, generate the indicator optimization plan through the associated constraint analysis module, and output the optimized indicator set; S4: Distribute the optimized indicator set to each business system through the version collaborative controller to establish a dynamic indicator update link.

2. The multi-dimensional business analysis indicator automatic adjustment and optimization management method according to claim 1, characterized in that: The S1 specifically includes: Decouple the features of data interfaces of multiple business systems through heterogeneous protocol parsers to generate unified data model templates; A dynamic sliding window mechanism is used to capture the time-varying characteristics of the data stream, and the data pattern mutation points are identified based on the LSTM-Attention model; Build a data fingerprint library including time and space dual-dimensional features, and eliminate semantic differences across business systems through fuzzy matching algorithms; The adaptive cleaning rule engine is triggered based on the pattern recognition results to generate a standardized data pool with version traceability identification, where the data items include: Metadata lineage map: identifies the original data source and conversion path; Dynamic confidence score: reflects the data quality level; Link anchor points across business systems.

3. The multi-dimensional business analysis indicator automatic adjustment and optimization management method according to claim 1, characterized in that: The S2 specifically includes: S21, build an indicator generation rule parser to convert the preset indicator rule set into a unified syntax tree structure. After parsing, each rule includes an operator sequence, data source mapping, and execution condition predicates, and establishes an executable and interpretable computing flow structure to support rule triggering and indicator derivation. S22, performing multimodal feature extraction processing on different types of data in the standardized data pool, integrating the features of each modality, and forming a unified indicator feature expression space; S23, builds a correlation map between indicators based on the indicator feature expression space, uses a multi-source heterogeneous similarity model to calculate the overall correlation strength between indicators, integrates feature space similarity, time series trajectory similarity and data source structure similarity, introduces a structural importance index for pruning the generated correlation edge set, and only retains the edge relationships corresponding to high similarity or structural key paths, forming a sparsely optimized indicator graph structure; S24, outputs an initial set of indicators that meet the validity conditions and retains their structural adjacency information in the association graph. This set serves as the input basis for subsequent indicator management, visualization, or adaptive control mechanisms.

4. The multi-dimensional business analysis index automatic adjustment and optimization management method according to claim 3 is characterized in that: The multimodal feature extraction process in S22 specifically includes: For numerical data, key features are extracted by combining statistical dimension reduction and signal transformation; For text data, the semantic expression vector is obtained by combining deep semantic modeling with keyword weights; For time series data, a feature channel based on time series dependency modeling is constructed to obtain the dynamic evolution characteristics of the context sequence.

5. The multi-dimensional business analysis index automatic adjustment and optimization management method according to claim 3, characterized in that: The S2 also includes dynamically assigning weights to the indicator set in combination with the scene context information, calculating the context importance score of each indicator through the similarity between the feature vector and the scene vector, executing the indicator validity verification process, and screening out unrepresentative or redundant indicators based on the dual conditions of the statistical deviation of the feature residual and the information entropy to obtain valid indicators.

6. The multi-dimensional business analysis index automatic adjustment and optimization management method according to claim 3, characterized in that: The initial indicator set is expressed as: ,in, represents the initial indicator set, As an effective indicator, Indicates that the indicator is in the indicator map A collection of adjacent related indicators, Indicator The context-aware weight value is used to measure the importance or priority of the indicator under the current scene characteristics. Represents the set of all valid indicators that are judged to be valid.

7. The multi-dimensional business analysis index automatic adjustment and optimization management method according to claim 1, characterized in that: The S3 specifically includes: S31, construct an indicator constraint relationship diagram to form an enhanced adjacency matrix representing the logical constraint relationship between indicators. By analyzing the rule description vector, a multi-layer perception network is used to convert it into constraint strength, and combined with the basic directional relationship between indicator pairs, a weighted adjacency matrix is ​​formed to characterize the directed constraint dependency between indicators. S32, based on the enhanced adjacency matrix, performs conflict path detection operations, uses a path synthesis algorithm to analyze whether there is a path conflict between any two indicators, and introduces a direction consistency judgment mechanism to identify path connections that potentially violate logical constraints; S33, after the conflict path detection operation, a multi-objective optimization function including the main objective and the constraint violation item is constructed. The multi-objective optimization function takes the deviation between the indicator function output and the target value as the main item, and introduces the penalty items of all the identified valid constraints. The constraint penalty intensity is dynamically adjusted with the violation frequency and the optimization process to form an adaptive control mechanism. S34, the weight variable of each indicator is gradient optimized based on the indicator constraint relationship graph. The gradient calculation takes into account the adjacency relationship of the indicators in the graph and propagates through the accumulated influence factors along the path, retaining the structural constraint dependency in the weight update process to ensure the topological consistency of the adjustment process.

8. The multi-dimensional business analysis index automatic adjustment and optimization management method according to claim 7, characterized in that: The S3 also includes a double check of stability and consistency of all updated indicators to determine whether the weight change is within the set range, and at the same time evaluate its logical consistency level under all associated constraints. When both conditions are met, the current indicator is included in the optimization indicator set and the optimization indicator set is output. Indicators that fail the verification will be marked as requiring recalculation.

9. The multi-dimensional business analysis index automatic adjustment and optimization management method according to claim 1, characterized in that: The S4 specifically includes: S41, generating an incremental update package: performing a difference analysis on the new and old indicator sets, calculating the change value of each indicator weight, forming a difference set consisting of an indicator identifier and a weight difference, and performing difference-driven coding compression on the difference set, specifically including hashing the indicator identifier and applying ZigZag coding to the weight difference, connecting the results after coding, and attaching an integrity check code to form an incremental update package, and calculating the compression rate by calculating the ratio of the total length after coding to the length of the original indicator set; S42, executing the improved transaction coordination protocol: the transaction update process includes pre-commit, compensation negotiation and final commit; S43, implement the gray release strategy: calculate the update priority of each business unit based on its current load level and criticality, and complete the batch division based on the priority. The number of business units included in each batch is related to the current system load pressure. The total number of batches is adjusted to achieve load balancing. During the batch update process, the indicator update tasks are deployed batch by batch in the order of priority from high to low; S44, build a health-driven rollback mechanism: monitor multiple key performance indicators, and calculate the overall health index of the business system based on their current values ​​and normal reference values. When it is detected that the overall health index continues to decline, and the rate of change exceeds the warning threshold and is lower than the minimum health tolerance value at the same time, it is determined to have entered a degraded state and automatically trigger the rollback operation of the indicator version.

10. The multi-dimensional business analysis index automatic adjustment and optimization management method according to claim 9, characterized in that: In the pre-commitment phase, a request message including an incremental update package and a timestamp is constructed, and the pre-update information is distributed and published to all business systems. After receiving the request, each business subsystem returns a signed confirmation response. After the coordinator collects multiple valid responses, it determines that the statutory response conditions are met. If the response is insufficient within the preset time, the compensation phase is entered, and an exponential backoff retry strategy is executed based on the time consumption and the number of failures. After the compensation attempt fails, the update process is terminated. If the response meets the conditions, the final submission phase is executed to construct a final transaction message including signature information and incremental data, and each business system performs a consistency check after receiving it.

Images