Data processing method and system based on ERP interface

Abstract

The invention discloses a data processing method and system based on an ERP (Enterprise Resource Planning) interface, and the method comprises the steps: obtaining time sequence features of historical data, carrying out fluctuation analysis and dynamic window clustering, dynamically generating a time window range, carrying out data mapping and grading according to a hierarchical structure, and generating a dynamic access control strategy; according to the dynamic access control strategy, double-target optimization iteration is carried out, an equilibrium threshold range is obtained, conflict prediction and dynamic locking are carried out, and a data processing result is obtained; according to a data processing result, performing fragment-copy collaborative adjustment to obtain a balance scheme, and performing knowledge graph modeling and priority scheduling to obtain an optimized balance scheme; and according to a time window range and the equilibrium threshold range, through collaborative optimization of a routing rule and reinforcement learning, obtaining an elastic scheduling result, and according to a fluctuation condition, carrying out multi-level early warning and strategy linkage to obtain a stable scheduling result. According to the invention, dynamic parameter adjustment can be realized so as to adapt to data fluctuation and business change.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular, to a data processing method and system based on an ERP interface. Background Art

[0002] In complex ERP data processing scenarios, the dynamic changes in the time distribution of data, access rights, and resource allocation often trigger multi-level chain effects. For example, the sudden increase in sales orders (such as during e-commerce promotion periods) may cause the data processing window to be overloaded, thereby triggering routing delays, concurrent conflicts, and resource contention. If the time window division strategy or the data segmentation threshold is not adjusted in time, it may further lead to system response delays, redundant data storage, and even interface call chaos.

[0003] In the prior art, ERP data processing mostly adopts methods such as fixed time window segmentation, role-based static access control (RBAC), experience-driven single-dimensional threshold setting, and static sharding-replica strategies. These methods rely on manual configuration and static rules and are difficult to adapt to data fluctuations and business changes. For example, fixed window division cannot cope with sudden data peaks, the RBAC model is difficult to meet dynamic business requirements, single-target threshold optimization leads to an imbalance between resources and efficiency, the static sharding strategy limits system elasticity, and exception handling relies on manual intervention and lacks a closed-loop feedback mechanism.

[0004] In summary, the prior art cannot adapt to data fluctuations and business changes, and there is a problem of insufficient dynamic adjustment ability. Summary of the Invention

[0005] The present invention provides a data processing method and system based on an ERP interface to achieve dynamic parameter adjustment so as to adapt to data fluctuations and business changes.

[0006] In a first aspect, to solve the above technical problems, the present invention provides a data processing method based on an ERP interface, including:

[0007] Obtain the time series characteristics of historical data, perform periodic fluctuation analysis and dynamic window clustering, and dynamically generate the time window range;

[0008] According to the time window range and the preset hierarchical structure, perform multi-granularity data mapping and sensitivity grading to generate a dynamic access control strategy;

[0009] According to the dynamic access control strategy, perform dual-objective optimization iteration to obtain an equilibrium threshold range;

[0010] According to the equilibrium threshold range, perform conflict prediction and dynamic locking to obtain a consistent data processing result;

[0011] Based on the data processing results, perform shard-replica collaborative adjustment to obtain a balancing scheme;

[0012] Based on the time window range and the equilibrium threshold range, through the collaborative optimization of multi-dimensional routing rules and reinforcement learning, obtain an elastic scheduling result;

[0013] Based on the dynamic access control policy and the balancing scheme, perform knowledge graph modeling and priority scheduling to obtain an optimized balancing scheme;

[0014] When the elastic scheduling result fluctuates abnormally, perform multi-level early warning and policy linkage to obtain a stable scheduling result.

[0015] As an optional implementation manner, the obtaining of the time series characteristics of historical data, performing periodic fluctuation analysis and dynamic window clustering, and dynamically generating a time window range includes:

[0016] Based on the time series characteristics of the historical data, perform hybrid periodic detection of Fourier transform and wavelet transform to obtain a fluctuation frequency spectrum;

[0017] Based on the fluctuation frequency spectrum, perform collaborative analysis of the sliding window algorithm and density clustering to obtain a set of candidate window intervals;

[0018] Based on the statistical distribution characteristics of the candidate window intervals, perform variance stability test and extreme value filtering processing to generate a time window range.

[0019] As an optional implementation manner, the performing of multi-granularity data mapping and sensitivity grading according to the time window range and a preset hierarchical structure to generate a dynamic access control policy includes:

[0020] Based on the time window range and the preset hierarchical structure, perform multi-granularity data weight allocation through the hierarchical clustering algorithm and the time decay factor to generate a spatio-temporal association mapping spectrum;

[0021] Based on the spatio-temporal association mapping spectrum, combine the entropy method and the random forest model to perform sensitivity probability evaluation and matrix construction to obtain a dynamic grading matrix;

[0022] Based on the dynamic grading matrix, perform policy weight iterative optimization through the ABAC policy model and the genetic algorithm to generate a dynamic access control policy integrating the time decay factor.

[0023] As an optional implementation manner, the performing of double-objective optimization iteration according to the dynamic access control policy to obtain an equilibrium threshold range includes:

[0024] According to the dynamic access control policy, a Pareto front solution set of conflict indicators is generated through the coupled modeling of the NSGA-II algorithm and constraint satisfaction technology;

[0025] According to the Pareto front solution set, a candidate threshold interval with optimal stability is obtained through the collaborative verification of moving window variance analysis and Monte Carlo simulation;

[0026] According to the candidate threshold interval, an equilibrium threshold range is obtained through a hybrid iterative mechanism of adaptive fuzzy control and gradient descent.

[0027] As an alternative implementation, the conflict prediction and dynamic locking are performed according to the equilibrium threshold range to obtain a consistent data processing result, including:

[0028] According to the equilibrium threshold range, a multi-dimensional conflict probability prediction model is generated through the coupled analysis of the LSTM network and association rule mining;

[0029] Based on the multi-dimensional conflict probability prediction model, a collaborative decision-making mechanism of dynamic confidence interval calculation and rule engine is adopted to trigger the adaptive locking of distributed lock granularity in real time;

[0030] According to the multi-dimensional conflict probability prediction model, a data processing result that meets the final consistency is generated through a hybrid verification strategy of multi-version snapshot comparison and rollback compensation.

[0031] As an alternative implementation, the sharding-replica collaborative adjustment is performed according to the data processing result to obtain a balancing scheme, including:

[0032] According to the data processing result, a dynamic sharding weight mapping table is generated through the coupled analysis of an adaptive scoring model and the consistent hashing algorithm;

[0033] Based on the dynamic sharding weight mapping table, the initial configuration of sharding granularity and replica distribution parameters is determined through the collaborative adjustment of moving window dynamic programming and heuristic rules;

[0034] According to the initial configuration, a sharding-replica balancing scheme that meets the requirements of load balancing and disaster tolerance is generated through the collaborative verification of cross-node delay simulation and data integrity verification.

[0035] As an alternative implementation, the elastic scheduling result is obtained through the collaborative optimization of multi-dimensional routing rules and reinforcement learning according to the time window range and the equilibrium threshold range, including:

[0036] According to the time window range and the equilibrium threshold range, a multi-constraint routing decision space including transmission delay, node load, and path reliability is constructed;

[0037] According to the multi-constraint routing decision space, a routing policy gradient optimization model is generated through a hybrid training mechanism of the Q-Learning algorithm and Dijkstra dynamic weights;

[0038] According to the collected real-time data and the routing policy gradient optimization model, dynamic inference and collaborative correction of the routing rule library are performed to obtain an elastic routing scheduling policy that meets the delay jitter constraint;

[0039] According to the elastic routing scheduling policy, stability tests of bandwidth utilization and packet loss rate are performed to generate a final elastic scheduling result.

[0040] As an optional implementation manner, the performing knowledge graph modeling and priority scheduling according to the dynamic access control policy and the balancing scheme to obtain an optimized balancing scheme includes:

[0041] According to the dynamic access control policy and the balancing scheme, a dynamically evolving knowledge graph is generated through the fusion modeling of graph neural networks and temporal feature extraction;

[0042] Based on the dynamically evolving knowledge graph, a multi-dimensional priority scoring model is constructed through the collaborative calculation of the improved PageRank algorithm and the deadline sensitivity coefficient;

[0043] According to the multi-dimensional priority scoring model, an optimal solution space for resource allocation is generated through a hybrid mechanism of pre-allocated buffer dynamic adjustment and preemptive scheduling;

[0044] According to the optimal solution space, a dual-channel verification of backpropagation correction and policy rollback is adopted to obtain an optimized balancing scheme that meets the load rebalancing constraint.

[0045] As an optional implementation manner, the performing multi-level early warning and policy linkage when the elastic scheduling result fluctuates abnormally to obtain a stable scheduling result includes:

[0046] According to the elastic scheduling result, multi-dimensional abnormal fluctuation indicators are generated through a hybrid detection of statistical process control charts and standard deviation analysis;

[0047] Based on the multi-dimensional abnormal fluctuation indicators, a three-level early warning signal is triggered through the collaborative determination of fuzzy membership functions and rule engines;

[0048] According to the three-level early warning signal, a combined emergency scheduling policy set is generated through a policy weight dynamic allocation algorithm and an incremental policy combination mechanism;

[0049] According to the combined emergency scheduling policy set, a stability evaluation of path switching success rate and recovery time is performed to obtain a stable scheduling result that meets the preset fault tolerance threshold.

[0050] In a second aspect, the present invention provides a data processing system based on an ERP interface, comprising:

[0051] A window clustering module, configured to obtain time series features of historical data, perform periodic fluctuation analysis and dynamic window clustering, and dynamically generate a time window range;

[0052] An access control module, configured to perform multi-granularity data mapping and sensitivity grading according to the time window range and a preset hierarchical structure, and generate a dynamic access control policy;

[0053] A dual-objective optimization module, configured to perform dual-objective optimization iteration according to the dynamic access control policy to obtain an equilibrium threshold range;

[0054] A dynamic locking module, configured to perform conflict prediction and dynamic locking according to the equilibrium threshold range to obtain a consistent data processing result;

[0055] A collaborative adjustment module, configured to perform sharding-replica collaborative adjustment according to the data processing result to obtain a balance scheme;

[0056] A collaborative optimization module, configured to perform collaborative optimization through multi-dimensional routing rules and reinforcement learning according to the time window range and the equilibrium threshold range to obtain an elastic scheduling result;

[0057] A graph optimization module, configured to perform knowledge graph modeling and priority scheduling according to the dynamic access control policy and the balance scheme to obtain an optimized balance scheme;

[0058] An early warning module, configured to perform multi-level early warning and policy linkage when the elastic scheduling result fluctuates abnormally to obtain a stable scheduling result.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] (1) By means of periodic analysis and clustering algorithms, the present invention adaptively adjusts the window range, solves the problem of uneven resource allocation, and improves the dynamic adjustment ability;

[0061] (2) By means of collaborative optimization of path allocation through reinforcement learning and multi-dimensional constraints, the present invention reduces the transmission delay and packet loss rate and improves the data transmission efficiency;

[0062] (3) Based on a closed-loop feedback mechanism, the present invention continuously optimizes the system performance through real-time monitoring and dynamic adjustment to ensure the high efficiency, security and stability of data processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1It is a schematic flowchart of a data processing method based on an ERP interface provided by an embodiment of the present invention;

[0064] Figure 2 It is a schematic structural diagram of a data processing system based on an ERP interface provided by an embodiment of the present invention. Specific embodiments

[0065] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0066] Referring to Figure 1 , the first embodiment of the present invention provides a data processing method based on an ERP interface, including the following steps:

[0067] S11. Obtain the time series characteristics of historical data, perform periodic fluctuation analysis and dynamic window clustering, and dynamically generate the time window range;

[0068] S12. According to the time window range and the preset hierarchical structure, perform multi-granularity data mapping and sensitivity grading to generate a dynamic access control policy;

[0069] S13. According to the dynamic access control policy, perform dual-objective optimization iteration to obtain an equilibrium threshold range;

[0070] S14. According to the equilibrium threshold range, perform conflict prediction and dynamic locking to obtain a consistent data processing result;

[0071] S15. According to the data processing result, perform sharding-replica collaborative adjustment to obtain a balancing scheme;

[0072] S16. According to the time window range and the equilibrium threshold range, through the collaborative optimization of multi-dimensional routing rules and reinforcement learning, obtain an elastic scheduling result;

[0073] S17. According to the dynamic access control policy and the balancing scheme, perform knowledge graph modeling and priority scheduling to obtain an optimized balancing scheme;

[0074] S18. When the elastic scheduling result fluctuates abnormally, perform multi-level early warning and policy linkage to obtain a stable scheduling result.

[0075] In step S11, obtaining the time series characteristics of historical data, performing periodic fluctuation analysis and dynamic window clustering, and dynamically generating the time window range includes:

[0076] Perform hybrid periodicity detection of Fourier transform and wavelet transform based on the time series characteristics of the historical data to obtain a fluctuation frequency spectrum;

[0077] Perform collaborative analysis of the sliding window algorithm and density clustering based on the fluctuation frequency spectrum to obtain a set of candidate window intervals;

[0078] Perform variance stability test and extreme value filtering processing based on the statistical distribution characteristics of the candidate window intervals to generate a time window range.

[0079] It should be noted that the dynamic generation operation of the time window range refers to extracting a time window range with significant periodicity and stability from historical data through steps such as hybrid periodicity detection, sliding window clustering, and statistical testing. This operation can help identify periodic patterns in the data in the data processing method based on the ERP interface, providing a scientific basis for subsequent feature extraction and model training. In the embodiments of the present invention, the dynamic generation of the time window range can efficiently extract a time window range with statistical significance through technical means such as hybrid analysis of Fourier transform and wavelet transform, collaborative optimization of sliding window and density clustering, and variance stability test, thereby providing reliable time dimension support for data processing.

[0080] Among them, the hybrid periodicity detection operation performs hybrid periodicity detection of Fourier transform and wavelet transform based on the time series characteristics of the historical data to obtain a fluctuation frequency spectrum. Specifically, first perform a fast Fourier transform (FFT) on the time series data X(t): Extract the main frequency component ω k , and its corresponding amplitude A k , to generate a preliminary frequency spectrum; secondly, perform a continuous wavelet transform (CWT) on the time series data using the Morlet wavelet basis function: [[ID=2i]] where a is the scale parameter, b is the translation parameter, ψ is the wavelet basis function, extract local frequency characteristics through wavelet transform, and supplement the global frequency information of the Fourier transform; finally, perform weighted fusion on the results of the Fourier transform and wavelet transform to generate a comprehensive fluctuation frequency spectrum: S(ω,a,b) = αF(ω)+(1-α)W(a,b), where α is the weight coefficient and is dynamically adjusted according to the data characteristics. The sliding window clustering and statistical testing operation The density clustering analysis operation performs collaborative analysis of the sliding window algorithm and density clustering based on the fluctuation frequency spectrum to obtain a set of candidate window intervals. Specifically, first perform sliding window partitioning, set the initial window size and sliding step length, and segment the time series data: X i = {X(t)|t∈[t i ,t i+W0]}, where t i represents the start time of the i-th window, and the calculation formula is t i = t0 + i·ΔW, i = 0, 1, 2,... ΔW represents the sliding step, that is, the interval between the start times of adjacent windows, and X i represents the data subset within the i-th time window, X(t) represents the original time series data, t is the time point, and W0 represents the window size, that is, the duration length of each time window; then, perform density-based clustering (DBSCAN) on the data points within each window. Specifically: First, calculate the Euclidean distance matrix between data points: D ij = ||X i - X j || 2 ; then identify core points, boundary points, and noise points according to the density threshold ∈ and the minimum number of points MinPts; finally, group the density-connected points into the same cluster to generate a set of candidate window interval sets C = {C k}. Statistical testing and filtering operations perform variance stability testing and extreme value filtering based on the statistical distribution characteristics of the candidate window intervals to generate the time window range. Specifically, first perform variance stability testing, calculate the variance of each candidate window interval and perform stability testing: where n k is the number of data points within the window, is the global variance. If exceeds the confidence interval, then eliminate this window; then perform extreme value filtering. For the window intervals that pass the variance test, perform extreme value filtering: First, calculate the upper and lower quartiles Q1 and Q3 of the data within the window; define the outlier range [Q1 - 1.5·IQR, Q3 + 1.5·IQR], where IQR = Q3 - Q1; finally, eliminate the window intervals containing outliers to generate the final time window range W = {W m}. Through the above steps, the dynamic generation operation of the time window range can efficiently extract time windows with significant periodicity and statistical stability, providing reliable time dimension support for data processing.

[0081] In step S12, the multi-granularity data mapping and sensitivity grading are performed according to the time window range and the preset hierarchical structure to generate a dynamic access control policy, including:

[0082] According to the time window range and the preset hierarchical structure, perform multi-granularity data weight allocation through the hierarchical clustering algorithm and the time decay factor to generate a spatio-temporal association mapping graph;

[0083] According to the spatio-temporal association mapping graph, the sensitivity probability evaluation and matrix construction are carried out by combining the entropy value method and the random forest model to obtain a dynamic grading matrix;

[0084] According to the dynamic grading matrix, the strategy weight iterative optimization is carried out through the ABAC strategy model and the genetic algorithm to generate a dynamic access control strategy integrating the time decay factor.

[0085] It should be noted that the generation operation of the dynamic access control strategy refers to constructing an access control strategy based on a time window and a hierarchical structure through steps such as multi-granularity data mapping, sensitivity grading, and strategy optimization. This operation can help achieve refined management of data access during the data processing process, ensuring the balance between data security and access efficiency. In the embodiments of the present invention, the generation of the dynamic access control strategy through the combination of hierarchical clustering and the time decay factor, the collaborative optimization of the entropy value method and the random forest model, and the iterative solution of the ABAC strategy model and the genetic algorithm can efficiently generate an access control strategy with strong adaptability and high security, thereby providing reliable data access guarantee for data processing.

[0086] Among them, the multi-granularity data mapping operation distributes the multi-granularity data weights through the hierarchical clustering algorithm and the time decay factor according to the time window range and the preset hierarchical structure to generate a spatio-temporal association mapping graph. Specifically, first, hierarchical clustering analysis is performed to perform hierarchical clustering on the data points within the time window and calculate the distance matrix between the data points: D ij =||X i -X j || 2 , and the Ward method is used for clustering and merging to generate a hierarchical structure tree T; then, according to the timestamp t i of the data, the time decay factor is calculated Among them, λ is the decay coefficient, and the time decay factor α i represents the weight decay degree of the i-th data point, t current represents the current timestamp, and t i represents the timestamp of the i-th data point; finally, combining the hierarchical clustering result and the time decay factor, the comprehensive weight w i =α i ·h i is calculated, where h iLet \(h\) be the hierarchical height. Integrate data points, clustering structures, and weight information into a graph structure, where the leaf nodes are the original data points, the internal nodes are the merge nodes of hierarchical clustering, the edges are the parent-child relationships of the hierarchical structure tree \(T\), and the node weights are the comprehensive weights of the data points, thereby generating the spatio-temporal correlation mapping graph \(G\). The sensitivity grading operation performs sensitivity probability evaluation and matrix construction according to the spatio-temporal correlation mapping graph, combining the entropy method and the random forest model, to obtain a dynamic grading matrix. Specifically, first calculate the information entropy of the data points where \(p\) ij is the probability distribution of feature \(j\), representing the probability that the feature value of data point \(i\) belongs to the \(j\)-th category, obtained by statistically counting the category frequencies, and \(k\) represents the total number of feature categories; secondly, use the random forest model to classify the sensitivity of the data points: construct a set of decision trees \(F = \{f\) k \}, each tree \(f\) k is generated by Bootstrap sampling, and calculate the sensitivity probability of the data points where \(f\) k (X i ) is the prediction output of the \(k\)-th tree for data point \(X\) i , and \(K\) is the total number of decision trees in the random forest; finally, combine the entropy method and the sensitivity classification results to construct a dynamic grading matrix \(M = [m\) ij , where \(m\) ij = w\) i ·P\) j , where \(w\) i is the comprehensive weight of data point \(i\), and \(P\) j is the sensitivity probability of belonging to the \(j\)-th category. The policy optimization operation performs iterative optimization of the policy weights through the ABAC policy model and the genetic algorithm according to the dynamic grading matrix, to generate a dynamic access control policy that incorporates a time decay factor. Specifically, first construct the ABAC policy model, and construct the ABAC policy model based on Attribute, Behavior, Condition, and Decision: define the attribute set \(A=\{a\) i \}, define the behavior set \(B = \{b\) j \}, define the condition set \(C=\{c\) k \}, define the decision set \(D = \{d\) l \}; the policy rules of the ABAC model require weights to quantify the influence degrees of different attributes, behaviors, and conditions on the decision, and use the genetic algorithm to perform iterative optimization of the policy weights: initialize the population \(P=\{p\) i \}, each individual \(p\) i represents a set of policy weights, and calculate the fitness function where \(s\) j is the policy score, and \(w\) jis the importance of the j-th policy rule, n is the total number of policy rules, and p i is a set of weight vectors. A new generation of population is generated through selection, crossover, and mutation operations, and the policy weights are iteratively optimized. Finally, according to the optimized policy weights, the product of the weights and the time decay factor is combined to obtain the comprehensive decision score of dynamic access control. The policies with scores exceeding the preset score threshold are selected, thereby generating a dynamic access control policy S = {s i}, where s i is the policy rule. Through the above steps, the generation operation of the dynamic access control policy can efficiently construct an access control policy with strong adaptability and high security, providing reliable data access guarantee for the data processing process.

[0087] In step S13, the double-objective optimization iteration is performed according to the dynamic access control policy to obtain the equilibrium threshold range, including:

[0088] According to the dynamic access control policy, a Pareto front solution set of conflict indicators is generated through the coupled modeling of the NSGA-II algorithm and constraint satisfaction technology;

[0089] According to the Pareto front solution set, the candidate threshold interval with the optimal stability is obtained through the collaborative verification of sliding window variance analysis and Monte Carlo simulation;

[0090] According to the candidate threshold interval, the equilibrium threshold range is obtained through the hybrid iterative mechanism of adaptive fuzzy control and gradient descent.

[0091] It should be noted that the generation operation of the equilibrium threshold range refers to constructing a threshold range that takes into account both security and response efficiency through steps such as multi-objective optimization, stability verification, and hybrid iteration. This operation can help achieve the dynamic adjustment of the access control policy in the data processing process, ensuring the robustness and efficiency of the system in a complex environment. In the embodiment of the present invention, the generation of the equilibrium threshold range can efficiently generate a threshold range with strong adaptability and high stability through the coupled modeling of the NSGA-II algorithm and constraint satisfaction technology, the collaborative verification of sliding window variance analysis and Monte Carlo simulation, and the hybrid iterative mechanism of adaptive fuzzy control and gradient descent, thereby providing reliable access control support for the data processing process.

[0092] Among them, the multi-objective optimization operation generates a Pareto front solution set of conflict indicators through the coupled modeling of the NSGA-II algorithm and constraint satisfaction technology according to the dynamic access control policy. Specifically, first, the NSGA-II algorithm is used to perform multi-objective optimization on the access control policy: define the objective functions f1 = -Security(S) (security) and f2 = -Efficiency(S) (response efficiency), and initialize the population P = {pi}, each individual p i represents a set of policy weights, and the Pareto front solution set F = {f is generated through non-dominated sorting and crowding degree calculation j}; Secondly, the Pareto front solution set is screened by combining constraint satisfaction technology: define the constraint condition c k (f) ≤ 0, where c k is the system constraint, and the solution set that satisfies the constraint condition is screened The stability verification operation is based on the Pareto front solution set, and the cooperative verification of sliding window variance analysis and Monte Carlo simulation is adopted to obtain the candidate threshold interval with the optimal stability. Specifically, first, perform sliding window variance analysis on the Pareto front solution set: set the sliding window size W and step size ΔW, and calculate the variance of the solution set within the window where is the mean value of the objective function values of the solution set within the window, f j is the objective function value of the jth solution, and the window interval I with the smallest variance is screened stable = [t start , t end ; Then, perform Monte Carlo simulation verification on the candidate window interval: generate a random sample X i ~N(μ, σ 2 ), calculate the stability index of the sample where N is the number of Monte Carlo simulations, s j (X i ) is the evaluation result of the jth sample X i , and the candidate threshold interval I with the optimal stability is screened candidate = [t a , t b . The hybrid iterative optimization operation obtains the balanced threshold range through the hybrid iterative mechanism of adaptive fuzzy control and gradient descent according to the candidate threshold interval. Specifically, first, use adaptive fuzzy control to preliminarily adjust the threshold range: define the fuzzy rule R k : IF(x1, x2) THEN(y k ), calculate the fuzzy output where w k is the activation weight of the kth rule, calculated through the membership function, y k is the threshold increment corresponding to the kth rule, and K is the total number of fuzzy rules; Then, use the gradient descent algorithm to iteratively optimize the fuzzy output: define the loss function where N is the number of training samples, y i is the fuzzy output value of the ith sample, y target is the desired optimal output value, and update the parameter where η is the learning rate, θk They are the threshold increase / decrease amount of the fuzzy rule and the membership function parameters; finally, an equilibrium threshold range is generated according to the optimization result. Through the above steps, the generation operation of the equilibrium threshold range can efficiently construct a threshold range that takes into account both security and response efficiency, providing reliable access control support for the data processing process.

[0093] In step S14, the conflict prediction and dynamic locking are performed according to the equilibrium threshold range to obtain a consistent data processing result, including:

[0094] According to the equilibrium threshold range, a multi-dimensional conflict probability prediction model is generated through the coupled analysis of the LSTM network and association rule mining;

[0095] Based on the multi-dimensional conflict probability prediction model, a collaborative decision-making mechanism of dynamic confidence interval calculation and rule engine is adopted to trigger the adaptive locking of the distributed lock granularity in real time;

[0096] According to the multi-dimensional conflict probability prediction model, a data processing result that meets the final consistency is generated through a hybrid verification strategy of multi-version snapshot comparison and rollback compensation.

[0097] It should be noted that the generation operation of the consistent data processing result refers to ensuring the consistency and reliability in the data processing process through steps such as conflict prediction, dynamic locking, and hybrid verification. This operation can help detect and dynamically adjust data access conflicts, ensuring data consistency and processing efficiency in a complex environment. In the embodiment of the present invention, the generation of the consistent data processing result can efficiently generate a data processing result that meets the final consistency through the coupled analysis of the LSTM network and association rule mining, the collaborative decision-making of dynamic confidence interval calculation and rule engine, and the hybrid verification strategy of multi-version snapshot comparison and rollback compensation, thereby providing reliable data processing support for the data processing process.

[0098] Among them, the conflict prediction operation generates a multi-dimensional conflict probability prediction model through the coupled analysis of the LSTM network and association rule mining according to the equilibrium threshold range. Specifically, first, the LSTM network is used to predict conflicts in time series data: define the input sequence X = [x1, x2,..., x n , x n is the observed value at the nth time point, calculate the LSTM cell state, and output the conflict probability P conflict = σ(W o · [h t-1 , x t + b o ), where σ is the sigmoid activation function, W o : output weight matrix, h t-1 : hidden state at the previous moment, xt : Input at the current moment, b o : Output bias term; then, use the Apriori algorithm to mine conflict association rules: calculate support Calculate confidence Generate a conflict association rule set R = {r k}, A ∪ B: The number of times events A and B occur simultaneously, N: Total number of samples; finally, combine the results of the LSTM network and association rule mining to generate a multi-dimensional conflict probability prediction model M = {P conflict , R}. The dynamic locking operation is based on the multi-dimensional conflict probability prediction model, and adopts a cooperative decision-making mechanism of dynamic confidence interval calculation and rule engine to trigger the adaptive locking of distributed lock granularity in real time. Specifically, first calculate the dynamic confidence interval according to the conflict probability P conflict μ P : Mean of the conflict probability P conflict , z: Critical value of the standard normal distribution, σ: Standard deviation of the conflict probability, n: Sample size; then use the rule engine to judge the conflict probability: Define the rule R k : IF (P conflict ≥θ k ) THEN (L k ), trigger the adaptive locking of the distributed lock granularity L k . The hybrid verification operation generates a data processing result that meets eventual consistency according to the multi-dimensional conflict probability prediction model through a hybrid verification strategy of multi-version snapshot comparison and rollback compensation. Specifically, first, perform a snapshot comparison on the data version: Generate a snapshot S i = {v j}, where v j is the data version, calculate the version difference ΔS = S i - S i-1 ; then perform rollback compensation on inconsistent data: Define the compensation function C(ΔS) = f compensate (ΔS), generate the final consistency result R final = S i + C(ΔS).

[0099] In step S15, the sharding-replica collaborative adjustment is performed according to the data processing result to obtain a balancing scheme, including:

[0100] Generate a dynamic sharding weight mapping table through the coupled analysis of the adaptive scoring model and the consistent hashing algorithm according to the data processing result;

[0101] ​Based on the dynamic shard weight mapping table, the initial configuration of the shard granularity and replica distribution parameters is determined by the collaborative adjustment of sliding window dynamic programming and heuristic rules;

[0102] According to the initial configuration, through the collaborative verification of cross-node latency simulation and data integrity verification, a shard-replica balance scheme that meets the requirements of load balancing and disaster tolerance is generated.

[0103] It should be noted that the generation operation of the shard-replica balance scheme refers to constructing a shard-replica distribution scheme that meets the requirements of load balancing and disaster tolerance through steps such as dynamic shard weight mapping, initial configuration adjustment, and collaborative verification. This operation can help achieve the efficiency and reliability of data storage and access, and ensure the performance stability and disaster tolerance of the system in a complex environment. In the embodiments of the present invention, the generation of the shard-replica balance scheme can efficiently generate a shard-replica balance scheme that meets the requirements of load balancing and disaster tolerance through the coupled analysis of the adaptive scoring model and the consistent hashing algorithm, the collaborative adjustment of the sliding window dynamic programming and heuristic rules, and the collaborative verification of the cross-node latency simulation and data integrity verification, thereby providing reliable data storage support for the data processing process.

[0104] Among them, the dynamic shard weight mapping operation generates a dynamic shard weight mapping table through the coupled analysis of the adaptive scoring model and the consistent hashing algorithm according to the data processing results. Specifically, first, the adaptive scoring model is used to score the data shards: define the scoring index S i = w1·Q i + w2·L i + w3·C i , where Q i is the query frequency, L i is the load pressure, C i is the data consistency, w1, w2, and w3 represent weight coefficients to balance the contributions of different indicators, and calculate the shard weight Among them, S i is the comprehensive score of the i-th data shard, reflecting its priority, and n is the total number of data shards; then, the consistent hashing algorithm is used to map the shards: construct a hash ring H = {h i}, where h i = hash(k i ), k i is the identification key of shard i, and map the shard to the node N j on the hash ring; finally, combining the results of the scoring model and the consistent hashing algorithm, generate a dynamic shard weight mapping table M = {W i , N j}。The initial configuration adjustment operation is based on the dynamic shard weight mapping table and uses the collaborative adjustment of sliding window dynamic programming and heuristic rules to determine the initial configuration of the shard granularity and replica distribution parameters. Specifically, first, the shard granularity is adjusted using sliding window dynamic programming: define the window size W and the step size ΔW, and calculate the optimal solution of the shard granularity within the window Then, the replica distribution parameters are adjusted using heuristic rules: define the rule R k :IF(W i ≥θ k )THEN(R k ), determine the replica distribution parameter P replica ={r j}, where W i is the shard weight, θ k is the critical value for triggering the replica adjustment rule, set through business requirements or historical data, and r j is the replica distribution parameter, including the number of replicas, the replica placement strategy (across racks, across availability zones), and the preferred deployment area; finally, combine the results of dynamic programming and heuristic rules to generate the initial configuration C={G i ,P replica}. The collaborative verification operation generates a shard-replica balance solution that meets the load balancing and disaster tolerance requirements through the collaborative verification of cross-node latency simulation and data integrity verification. Specifically, first, the latency simulation of the shard-replica distribution is performed: define the latency model D ij =d base +k ij ·l ij , and calculate the average latency where D ij is the access latency from shard i to replica j, which consists of the basic latency and the network transmission latency, d base represents the basic latency, k ij represents the network transmission coefficient, reflecting the link quality between shard i and replica j, l ij represents the data volume from shard i to replica j, represents the average latency of all shard-replica pairs, and N is the total number of shard-replica pairs; then, the integrity verification of the replica data is performed: define the verification function V(R)=f verify (R), and calculate the integrity score where V(R) is the integrity verification result of replica R, f verify represents the verification function, S integrity represents the verification passing rate of all replicas, M is the total number of replicas, and V j is the jth replica; finally, generate the shard-replica balance solution according to the results of latency simulation and integrity verification

[0105] In step S16, obtaining an elastic scheduling result through collaborative optimization of multi-dimensional routing rules and reinforcement learning according to the time window range and the equilibrium threshold range includes:

[0106] Constructing a multi-constraint routing decision space including transmission delay, node load, and path reliability according to the time window range and the equilibrium threshold range;

[0107] Generating a routing policy gradient optimization model through a hybrid training mechanism of the Q-Learning algorithm and Dijkstra dynamic weights according to the multi-constraint routing decision space;

[0108] Performing collaborative correction of dynamic inference and a routing rule library according to the collected real-time data and the routing policy gradient optimization model to obtain an elastic routing scheduling policy that satisfies the delay jitter constraint;

[0109] Performing stability tests on bandwidth utilization and packet loss rate according to the elastic routing scheduling policy to generate a final elastic scheduling result.

[0110] It should be noted that the operation of generating the elastic scheduling result refers to constructing an elastic routing scheduling policy that satisfies the delay jitter constraint through steps such as constructing a multi-constraint routing decision space, optimizing the routing policy gradient, and correcting dynamic inference. This operation can help achieve the efficiency and stability of data transmission, and ensure the performance optimization and resource utilization of the system in a complex environment. In the embodiments of the present invention, the generation of the elastic scheduling result through the collaborative optimization of multi-dimensional routing rules and reinforcement learning, the hybrid training mechanism of the Q-Learning algorithm and Dijkstra dynamic weights, and the stability tests of bandwidth utilization and packet loss rate can efficiently generate an elastic scheduling result that satisfies the delay jitter constraint, thereby providing reliable data transmission support for the data processing process.

[0111] Among them, the operation of constructing the multi-constraint routing decision space constructs a multi-constraint routing decision space including transmission delay, node load, and path reliability according to the time window range and the equilibrium threshold range. Specifically, first define the transmission delay model D ij = d base + k ij · l ij , where d base is the basic delay, k ij is the path coefficient, and l ij is the path length; then, define the node load model where Q i is the node queue length, and C i is the node capacity; secondly, define the path reliability model where rk is the reliability of the path segment, n represents the total number of path segments, which is determined by the actual network topology; finally, combining the transmission delay, node load, and path reliability model, a multi-constraint routing decision space S = {D ij , L i , R ij} is generated. The routing policy gradient optimization operation generates a routing policy gradient optimization model according to the multi-constraint routing decision space through a hybrid training mechanism of the Q-Learning algorithm and Dijkstra dynamic weights. Specifically, first, the Q-Learning algorithm is used for routing policy optimization: define the state s t and the action a t , and update the Q value where r t represents the feedback after executing the action a t , γ represents the discount factor, which measures the importance of future rewards and is set to 0.9; then, the Dijkstra algorithm is used for dynamic weight adjustment: define the dynamic weight w[[ID=;18]] ij = α1·D ij + α2·L i + α3·R ij , and calculate the shortest path where α1, α2, and α3 represent weight coefficients, which balance the importance of the delay D ij , the load L i , and the link reliability R ij , and n represents the link length; finally, combining the results of the Q-Learning algorithm and Dijkstra dynamic weight adjustment, a routing policy gradient optimization model M = {Q t , P min} is generated. The dynamic inference and routing rule base correction operation performs dynamic inference and collaborative correction of the routing rule base according to the collected real-time data and the routing policy gradient optimization model to obtain an elastic routing scheduling policy that meets the delay jitter constraint. Specifically, first, a dynamic inference algorithm is used to process the real-time data: define the inference function F(X) = f inference (X), and calculate the inference result Y t = F(X t ); then, correct the routing rule base according to the inference result: define the correction rule R k : IF(Y t ≥ θ k ) THEN(R k ), and update the routing rule base R = {R k}; finally, combining the results of dynamic inference and routing rule base correction, an elastic routing scheduling policy S = {Y t, R}. The stability test and elastic scheduling result generation operation perform a stability test on the bandwidth utilization rate and packet loss rate according to the elastic routing scheduling policy, and generate the final elastic scheduling result. Specifically, first calculate the bandwidth utilization rate Then, calculate the packet loss rate where B used represents the actual amount of data transmitted on the current link, obtained from the router / switch through SNMP, NetFlow, sFlow, and B total represents the theoretical maximum capacity of the link, set by the network administrator, and N loss represents the number of data packets lost during the transmission, obtained from the packet loss counter of the router / switch, and N total represents the total number of data packets attempted to be transmitted; finally, according to the test results of the bandwidth utilization rate and packet loss rate, generate the final elastic scheduling result R = {U bw , P loss}. Through the above steps, the generation operation of the elastic scheduling result can efficiently construct an elastic routing scheduling policy that meets the delay jitter constraint, providing reliable data transmission support for the data processing process.

[0112] In step S17, the knowledge graph modeling and priority scheduling are performed according to the dynamic access control policy and the balancing scheme to obtain an optimized balancing scheme, including:

[0113] According to the dynamic access control policy and the balancing scheme, generate a dynamically evolving knowledge graph through the fusion modeling of graph neural networks and temporal feature extraction;

[0114] Based on the dynamically evolving knowledge graph, construct a multi-dimensional priority scoring model by the collaborative calculation of the PageRank improvement algorithm and the deadline sensitivity coefficient;

[0115] According to the multi-dimensional priority scoring model, generate the optimal solution space of resource allocation through the hybrid mechanism of pre-allocated buffer dynamic adjustment and preemptive scheduling;

[0116] According to the optimal solution space, obtain an optimized balancing scheme that meets the load rebalancing constraint through the dual-channel verification of backpropagation correction and policy rollback.

[0117] It should be noted that the generation operation of the optimization balance scheme refers to constructing an optimization balance scheme that meets the load rebalancing constraints through steps such as knowledge graph modeling, construction of a priority scoring model, and optimization of resource allocation. This operation can help achieve the efficiency and fairness of resource allocation, ensuring the performance optimization and load balancing of the system in a complex environment. In the embodiments of the present invention, the generation of the optimization balance scheme can efficiently generate an optimization balance scheme that meets the load rebalancing constraints through the fusion modeling of graph neural networks and temporal feature extraction, the collaborative calculation of the improved PageRank algorithm and the deadline sensitivity coefficient, and the hybrid mechanism of dynamic adjustment of the pre-allocated buffer and preemptive scheduling, thereby providing reliable resource allocation support for the data processing process.

[0118] Among them, the dynamic evolution knowledge graph modeling operation generates a dynamic evolution knowledge graph through the fusion modeling of graph neural networks and temporal feature extraction according to the dynamic access control policy and the balance scheme. Specifically, first, a graph neural network (GNN) is used to model the knowledge graph: define the node feature h i and the edge feature e ij , calculate the node update Then, a temporal feature extraction algorithm is used to extract the dynamic evolution features: define the temporal feature f t =f extract (X t ), calculate the dynamic evolution feature Among them, represents the feature vector of node j at time step t, N(i) represents the neighbor set of node i, W ij represents the weight matrix of edge (i,j), b i represents the bias term of node i, σ represents the activation function sigmoid, X t represents the input data at time step t, f extract represents the feature extraction function, which is extracted by using a sliding window to statistically calculate the mean and variance, f t represents the temporal feature of node i at time t, n represents the total number of nodes; finally, combining the results of graph neural network and temporal feature extraction, generate the dynamic evolution knowledge graph G={h i ,e ij ,F t}. The construction operation of the multi-dimensional priority scoring model constructs a multi-dimensional priority scoring model based on the dynamic evolution knowledge graph by using the collaborative calculation of the improved PageRank algorithm and the deadline sensitivity coefficient. Specifically, first, the improved PageRank algorithm is used to calculate the node priority: define the PageRank value Calculate the improved PageRank value PR′ i =PR i ·w i; Then calculate the deadline sensitivity coefficient where PR i represents the classical PageRank value of node i, D represents the damping factor (set to 0.85), N represents the total number of nodes, N(i) represents the set of in-link neighbors of node i, and w i represents the local weight of node i, t i represents the evaluation moment of the task or node, t deadline represents the final deadline by which the task must be completed, k represents the steepness of the control function near the deadline; finally, combining the results of the PageRank improvement algorithm and the deadline sensitivity coefficient, a multi-dimensional priority scoring model M = {PR′ i , α i} is generated. The optimal solution space generation operation for resource allocation generates the optimal solution space for resource allocation according to the multi-dimensional priority scoring model through a hybrid mechanism of pre-allocated buffer dynamic adjustment and preemptive scheduling. Specifically, first, a pre-allocated buffer dynamic adjustment mechanism is adopted: define the buffer size B i = B base + k i ·w i , dynamically adjust the buffer B i = B i + ΔB; then adopt a preemptive scheduling mechanism for resource allocation: define the preemption rule R k : IF(P i ≥ θ k ) THEN(S k ), generate the preemption scheduling result S k = {s j}, where B base represents the basic buffer size, k i represents the elasticity coefficient, controlling the expansion amplitude of the buffer with the task weight w i , ΔB represents the dynamic adjustment amount, increasing or decreasing the buffer according to the real-time load, P i represents the priority of task i, θ k represents the preemption threshold, hierarchically set as θ1 = 0.7, θ2 = 0.9, and S k represents the set of preemption actions, the list of preempted tasks or resource release instructions; finally, combining the results of the pre-allocated buffer dynamic adjustment and the preemptive scheduling mechanism, an optimal solution space O = {B i , S k} is generated. The dual-channel verification and optimization balance scheme generation operation obtains an optimized balance scheme that satisfies the load rebalancing constraint by using dual-channel verification of backpropagation correction and policy rollback according to the optimal solution space. Specifically, first, use the backpropagation algorithm to correct the optimal solution space: define the loss function Update parameters where N is the number of training samples, y i is the fuzzy output value of the i-th sample, y target is the expected optimal output value, η is the learning rate, and θ k is the k-th model parameter to be optimized; then a policy rollback mechanism is used to verify the correction result: define a rollback rule R k : IF (L i ≥ θ k ) THEN (R k ), perform the rollback operation R k = R k - ΔR; finally, combining the results of backpropagation correction and policy rollback verification, generate an optimized balance scheme B = {O i , R k}, where L i represents the loss value after the i-th correction, θ k represents the rollback trigger threshold, and when the loss exceeds this value, the correction is determined to fail. ΔR represents the rollback step size, and R k represents the rollback version number or parameter snapshot.

[0119] In step S18, when the elastic scheduling result fluctuates abnormally, perform multi-level early warning and policy linkage to obtain a stable scheduling result, including:

[0120] According to the elastic scheduling result, generate a multi-dimensional abnormal fluctuation index through a hybrid detection of a statistical process control chart and standard deviation analysis;

[0121] Based on the multi-dimensional abnormal fluctuation index, trigger a three-level early warning signal by the collaborative determination of a fuzzy membership function and a rule engine;

[0122] According to the three-level early warning signal, generate a combined emergency scheduling policy set through a policy weight dynamic allocation algorithm and an incremental policy combination mechanism;

[0123] According to the combined emergency scheduling policy set, perform a stability assessment of the path switching success rate and recovery time to obtain a stable scheduling result that meets the preset fault tolerance threshold.

[0124] It should be noted that the operation of generating a stable scheduling result refers to constructing a stable scheduling result that meets the fault tolerance threshold through steps such as abnormal fluctuation detection, multi-level early warning triggering, and emergency strategy linkage. This operation can help achieve the abnormal detection and emergency response of the scheduling system, ensuring the stability and fault tolerance of the system in a complex environment. In the embodiments of the present invention, the generation of a stable scheduling result can efficiently generate a stable scheduling result that meets the fault tolerance threshold through a hybrid detection of a statistical process control chart and standard deviation analysis, a collaborative determination of a fuzzy membership function and a rule engine, and a dynamic allocation algorithm of policy weights and an incremental policy combination mechanism.

[0125] Among them, the abnormal fluctuation detection operation generates multi-dimensional abnormal fluctuation indicators through a hybrid detection of a statistical process control chart and standard deviation analysis based on the elastic scheduling result. Specifically, first, a statistical process control chart (SPC) is used to detect the scheduling result: define the upper control limit Define the lower control limit Detect abnormal points Then, standard deviation analysis is used to quantify the abnormal fluctuation: calculate the standard deviation Generate abnormal fluctuation indicators Among them, represents the sample mean, σ i represents the sample standard deviation, k represents the control limit coefficient, which determines the looseness of the control range and is set to 3, LCL i , UCL i represent the upper and lower control limits, the boundaries of normal fluctuations, X j represents the jth historical scheduling result, N represents the sample size, and the data within the sliding window is taken. σ base represents the reference standard deviation, which is obtained through the data statistics of the long-term stable operation stage; finally, combining the results of the statistical process control chart and standard deviation analysis, a multi-dimensional abnormal fluctuation indicator M = {UCL i , LCL i , I abnormal} is generated. The multi-level early warning triggering operation triggers a three-level early warning signal based on the multi-dimensional abnormal fluctuation indicator through the collaborative determination of a fuzzy membership function and a rule engine. Specifically, first, a fuzzy membership function is used to quantify the abnormal fluctuation: define the membership function Calculate the fuzzy membership μ i ′ = μ i ·w i ; then, a rule engine is used to determine the abnormal fluctuation: define the rule R k : IF(μ i ≥θ k ) THEN(S k ), trigger the three-level early warning signal S = {S k}, where, Xi Denote the current abnormal fluctuation value, θ i Denote the membership center threshold, the standard value for determining abnormality, set to 1.5, k denotes the membership slope, μ i Denote the original membership, μ i ′ denotes the weighted membership, w i Denote the weight coefficient, adjusting the importance of the membership, θ k Denote the rule threshold, the lowest membership for triggering an early warning (θ k = 0.6), S k Denote the early warning action, corresponding to response measures at different levels. The emergency strategy linkage operation generates a combined emergency dispatching strategy set through the strategy weight dynamic allocation algorithm and the incremental strategy combination mechanism according to the three-level early warning signal. Specifically, first, the strategy weight dynamic allocation algorithm is used to allocate weights to the emergency strategies: define the weight Calculate the strategy weight w i ′ = w i ·α i ; Then, the incremental strategy combination mechanism is used to generate the emergency dispatching strategy set: define the strategy combination Generate the combined emergency dispatching strategy set C = {C k}, where, μ i Denote the fuzzy membership of strategy i, α i Denote the time decay factor, set to 0.01, w i Denote the normalized weight, S i Denote the i-th basic strategy, C k Denote the combined strategy, a linear combination of multiple basic strategies, C denotes the emergency strategy set, including all possible combined strategies. The stability evaluation and steady dispatching result generation operation perform the stability evaluation of the path switching success rate and the recovery time according to the combined emergency dispatching strategy set, and obtain the steady dispatching result that meets the preset fault tolerance threshold. Specifically, first calculate the path switching success rate Then calculate the recovery time Finally, according to the evaluation results of the path switching success rate and the recovery time, generate the steady dispatching result R = {P switch , T recovery}, where N success Denote the number of successful path switches within the specified time, N total Denote the total number of switching attempts, including successful and failed switching operations, P switch Denote the switching success rate, t i Denote the time taken from the occurrence of the fault to the complete recovery of the service, n denotes the total number of faults, the number of fault events within the statistical period, T recovery Denote the total recovery time, reflecting the disaster tolerance efficiency of the system.

[0126] Reference Figure 2 The second embodiment of the present invention provides a data processing system based on an ERP interface, comprising:

[0127] Window clustering module, used to obtain the time series characteristics of historical data, perform periodic fluctuation analysis and dynamic window clustering, and dynamically generate time window ranges;

[0128] An access control module is used to perform multi-granularity data mapping and sensitivity classification according to the time window range and the preset hierarchical structure, and generate a dynamic access control policy;

[0129] A dual-objective optimization module, configured to perform dual-objective optimization iterations according to the dynamic access control policy to obtain a balanced threshold range;

[0130] A dynamic locking module, configured to perform conflict prediction and dynamic locking according to the equilibrium threshold range to obtain consistent data processing results;

[0131] A collaborative adjustment module, configured to perform shard-replica collaborative adjustment based on the data processing results to obtain a balancing solution;

[0132] A collaborative optimization module, configured to obtain a flexible scheduling result by collaboratively optimizing multi-dimensional routing rules and reinforcement learning according to the time window range and the balance threshold range;

[0133] A graph optimization module, configured to perform knowledge graph modeling and priority scheduling according to the dynamic access control strategy and the balancing solution to obtain an optimized balancing solution;

[0134] The early warning module is used to perform multi-level early warning and strategy linkage when the elastic scheduling result fluctuates abnormally to obtain a stable scheduling result.

[0135] It should be noted that the data processing system based on the ERP interface provided in an embodiment of the present invention is used to execute all the process steps of the data processing method based on the ERP interface in the above embodiment. The working principles and beneficial effects of the two correspond one to one, so they will not be repeated here.

[0136] The embodiment of the present invention further provides a terminal device. The terminal device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a data processing program based on an ERP interface. When the processor executes the computer program, the steps in the above-mentioned embodiments of the data processing method based on an ERP interface are implemented, such as Figure 1 Alternatively, the processor implements the functions of the modules / units in the above-mentioned system embodiments when executing the computer program.

[0137] Exemplarily, the computer program may be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the terminal device.

[0138] The terminal device may be a computing device such as a desktop computer, a notebook, a palm computer, and a smart tablet. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the terminal device and do not constitute a limitation on the terminal device. It may include more or fewer components than the above, or combine certain components, or different components. For example, the terminal device may further include input / output devices, network access devices, a bus, etc.

[0139] The so-called processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the terminal device and connects various parts of the entire terminal device through various interfaces and lines.

[0140] The memory can be used to store the computer program and / or modules. By running or executing the computer program and / or modules stored in the memory, and invoking the data stored in the memory, the processor realizes various functions of the terminal device. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0141] Among them, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-mentioned embodiment methods of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, an executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or system, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium that can carry the computer program code. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0142] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the attached drawings of the system embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0143] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. In particular, for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A data processing method based on an ERP interface, characterized in that Including: Obtain the time series features of historical data, conduct periodic fluctuation analysis and dynamic window clustering, and dynamically generate the time window range; According to the time window range and the preset hierarchical structure, conduct multi-granularity data mapping and sensitivity grading to generate a dynamic access control policy; According to the dynamic access control policy, conduct dual-objective optimization iteration to obtain the equilibrium threshold range; According to the equilibrium threshold range, conduct conflict prediction and dynamic locking to obtain a consistent data processing result; According to the data processing result, conduct sharding-replica collaborative adjustment to obtain a balancing scheme; According to the time window range and the equilibrium threshold range, through the collaborative optimization of multi-dimensional routing rules and reinforcement learning, obtain an elastic scheduling result; According to the dynamic access control policy and the balancing scheme, conduct knowledge graph modeling and priority scheduling to obtain an optimized balancing scheme; When the elastic scheduling result fluctuates abnormally, conduct multi-level early warning and policy linkage to obtain a stable scheduling result.

2. The data processing method based on the ERP interface according to claim 1, wherein The obtaining the time series features of historical data, conducting periodic fluctuation analysis and dynamic window clustering, and dynamically generating the time window range includes: According to the time series features of the historical data, conduct hybrid periodic detection of Fourier transform and wavelet transform to obtain a fluctuation frequency spectrum; According to the fluctuation frequency spectrum, conduct collaborative analysis of the sliding window algorithm and density clustering to obtain a set of candidate window intervals; According to the statistical distribution characteristics of the candidate window intervals, conduct variance stability test and extreme value filtering processing to generate the time window range.

3. The data processing method based on an ERP interface according to claim 1, wherein The conducting multi-granularity data mapping and sensitivity grading according to the time window range and the preset hierarchical structure to generate a dynamic access control policy includes: According to the time window range and the preset hierarchical structure, conduct multi-granularity data weight allocation through the hierarchical clustering algorithm and the time decay factor to generate a spatio-temporal association mapping spectrum; According to the spatio-temporal association mapping spectrum, combine the entropy method and the random forest model to conduct sensitivity probability evaluation and matrix construction to obtain a dynamic grading matrix; According to the dynamic grading matrix, conduct policy weight iterative optimization through the ABAC policy model and the genetic algorithm to generate a dynamic access control policy integrating the time decay factor.

4. The data processing method based on an ERP interface according to claim 1, wherein The conducting dual-objective optimization iteration according to the dynamic access control policy to obtain the equilibrium threshold range includes: According to the dynamic access control policy, generate a Pareto front solution set of conflict indicators through the coupled modeling of the NSGA-II algorithm and the constraint satisfaction technology; According to the Pareto front solution set, conduct collaborative verification of sliding window variance analysis and Monte Carlo simulation to obtain a candidate threshold interval with optimal stability; According to the candidate threshold interval, obtain the equilibrium threshold range through the hybrid iteration mechanism of adaptive fuzzy control and gradient descent.

5. The data processing method based on the ERP interface according to claim 1, characterized in that The conducting conflict prediction and dynamic locking according to the equilibrium threshold range to obtain a consistent data processing result includes: According to the equilibrium threshold range, generate a multi-dimensional conflict probability prediction model through the coupled analysis of the LSTM network and association rule mining; Based on the multi-dimensional conflict probability prediction model, a collaborative decision-making mechanism combining dynamic confidence interval calculation and rule engine is adopted to trigger the adaptive locking of distributed lock granularity in real time; According to the multi-dimensional conflict probability prediction model, a hybrid verification strategy of multi-version snapshot comparison and rollback compensation is used to generate a data processing result that meets eventual consistency.

6. The data processing method based on an ERP interface according to claim 1, wherein According to the data processing result, sharding-replica collaborative adjustment is performed to obtain a balancing scheme, including: According to the data processing result, a dynamic sharding weight mapping table is generated through the coupled analysis of an adaptive scoring model and the consistent hashing algorithm; Based on the dynamic sharding weight mapping table, a collaborative adjustment of sliding window dynamic programming and heuristic rules is adopted to determine the initial configuration of sharding granularity and replica distribution parameters; According to the initial configuration, a sharding-replica balancing scheme that meets the requirements of load balancing and disaster tolerance is generated through the collaborative verification of cross-node delay simulation and data integrity verification.

7. The data processing method based on an ERP interface according to claim 1, wherein According to the time window range and the equilibrium threshold range, a flexible scheduling result is obtained through the collaborative optimization of multi-dimensional routing rules and reinforcement learning, including: According to the time window range and the equilibrium threshold range, a multi-constraint routing decision space including transmission delay, node load, and path reliability is constructed; According to the multi-constraint routing decision space, a routing policy gradient optimization model is generated through a hybrid training mechanism of the Q-Learning algorithm and Dijkstra dynamic weights; According to the collected real-time data and the routing policy gradient optimization model, dynamic reasoning and collaborative correction of the routing rule library are performed to obtain a flexible routing scheduling strategy that meets the delay jitter constraint; According to the flexible routing scheduling strategy, stability tests of bandwidth utilization and packet loss rate are performed to generate the final flexible scheduling result.

8. The data processing method based on the ERP interface according to claim 1, characterized in that According to the dynamic access control policy and the balancing scheme, knowledge graph modeling and priority scheduling are performed to obtain an optimized balancing scheme, including: According to the dynamic access control policy and the balancing scheme, a dynamically evolving knowledge graph is generated through the fusion modeling of graph neural networks and temporal feature extraction; Based on the dynamically evolving knowledge graph, a multi-dimensional priority scoring model is constructed through the collaborative calculation of the improved PageRank algorithm and the deadline sensitivity coefficient; According to the multi-dimensional priority scoring model, an optimal solution space for resource allocation is generated through a hybrid mechanism of pre-allocated buffer dynamic adjustment and preemptive scheduling; According to the optimal solution space, a dual-channel verification of backpropagation correction and policy rollback is adopted to obtain an optimized balancing scheme that meets the load rebalancing constraint.

9. The data processing method based on an ERP interface according to claim 1, characterized in that When the flexible scheduling result fluctuates abnormally, multi-level early warning and policy linkage are performed to obtain a stable scheduling result, including: According to the flexible scheduling result, multi-dimensional abnormal fluctuation indicators are generated through the hybrid detection of statistical process control charts and standard deviation analysis; Based on the multi-dimensional abnormal fluctuation indicators, a three-level early warning signal is triggered through the collaborative decision-making of fuzzy membership functions and rule engines; According to the three-level early warning signal, a combined emergency scheduling policy set is generated through a policy weight dynamic allocation algorithm and an incremental policy combination mechanism; Based on the combined emergency scheduling strategy set, perform stability evaluations of the path switching success rate and recovery time to obtain a stable scheduling result that meets the preset fault tolerance threshold.

10. A data processing system based on an ERP interface, characterized in that, Including: A window clustering module, which is used to obtain the time series characteristics of historical data, perform periodic fluctuation analysis and dynamic window clustering, and dynamically generate the time window range; An access control module, which is used to perform multi-granularity data mapping and sensitivity grading according to the time window range and the preset hierarchical structure, and generate a dynamic access control strategy; A dual-objective optimization module, which is used to perform dual-objective optimization iterations according to the dynamic access control strategy to obtain an equilibrium threshold range; A dynamic locking module, which is used to perform conflict prediction and dynamic locking according to the equilibrium threshold range to obtain a consistent data processing result; A collaborative adjustment module, which is used to perform sharding-replica collaborative adjustment according to the data processing result to obtain a balancing scheme; A collaborative optimization module, which is used to obtain an elastic scheduling result through the collaborative optimization of multi-dimensional routing rules and reinforcement learning according to the time window range and the equilibrium threshold range; A graph optimization module, which is used to perform knowledge graph modeling and priority scheduling according to the dynamic access control strategy and the balancing scheme to obtain an optimized balancing scheme; An early warning module, which is used to perform multi-level early warning and policy linkage when the elastic scheduling result fluctuates abnormally to obtain a stable scheduling result.

Images