Enterprise digitization system and method based on SAAS micro-service architecture

Abstract

The application discloses an enterprise digitization system and method based on a SAAS micro-service architecture, relates to the technical field of enterprise digitization, and comprises the following steps: S1, collecting multi-tenant micro-service calling data and configuration information, and generating service path basic data; S2, constructing a jump path tensor structure to represent multi-hop dependence and context relationship between services; S3, constructing a path modeling network containing path attention and a gating mechanism, and extracting key path features; S4, setting a structure optimization function with response delay, path accuracy and model complexity as targets; S5, applying a spotted hyena optimization algorithm to globally optimize network structure parameters; and S6, outputting optimal structure and modeling parameters, and realizing system structure deployment and service scheduling optimization. The application realizes accurate modeling and efficient optimization deployment of cross-tenant service paths, and significantly improves system response performance and structure stability.

Description

Technical Field

[0001] This invention relates to the field of enterprise digital technology, and in particular to enterprise digital systems and methods based on SaaS microservice architecture. Background Technology

[0002] As enterprises continue their digital transformation, more and more software systems are adopting SaaS architecture to meet the needs of flexible deployment, elastic scaling, and multi-tenancy management. Under this architecture, microservices technology has become a crucial foundation for supporting complex business processes. It breaks down traditional monolithic applications into multiple independently running small services, enabling enterprises to combine and dynamically adjust functional modules as needed. In a multi-tenant environment, the business processes of different enterprise users vary significantly. The system needs to implement personalized service path scheduling and process optimization based on the behavioral patterns of different tenants. Therefore, path modeling and intelligent scheduling become key issues in platform construction.

[0003] Current mainstream approaches to path modeling often rely on fixed-structure deep learning models or rule-based sequence analysis methods, such as recurrent neural networks or graph-based algorithms. While these methods can achieve certain results in some scenarios, they exhibit significant limitations when dealing with highly complex, multi-hop, and non-linear service call relationships. On the one hand, most existing models depend on pre-defined parameter structures, lacking flexibility and unable to adaptively adjust the structure based on tenant behavior differences. On the other hand, cross-node hop relationships in the path structure are often simplified to continuous dependencies, failing to effectively reflect the true complexity of service calls. Furthermore, most models focus only on a single performance metric during optimization, lacking a comprehensive trade-off between prediction accuracy, structural coherence, and anomaly detection capabilities.

[0004] Another common problem lies in parameter optimization methods. Most current systems employ traditional gradient descent for model tuning. However, this approach is prone to getting stuck in local optima for models with complex structures or high parameter space dimensions, resulting in unstable optimization performance. Furthermore, in actual deployments, the service paths of different tenants vary significantly. Without dynamic optimization and multi-objective control of structural parameters, it is difficult to meet the platform's need for a balance between generality and personalization.

[0005] Therefore, how to construct a path modeling scheme that is highly expressive and possesses structural self-optimization capabilities for multi-tenant scenarios is a problem that current enterprise-level SaaS platforms urgently need to solve. This invention addresses this practical need by proposing a technical solution that combines tensor structure modeling with evolutionary optimization algorithms. By introducing jump relation representation, path attention mechanisms, and context gating structures, the accuracy and adaptability of path modeling are improved. Simultaneously, the spotted hyena optimization algorithm is used to perform global search and multi-objective evaluation of structural parameters, enabling the model to maintain stable performance across different tenant scenarios. This approach has strong practical value in solving problems such as structural rigidity and difficulties in parameter tuning. Summary of the Invention

[0006] One objective of this invention is to propose a method for SaaS microservice path modeling and parameter optimization based on the Spotted Hyena optimization algorithm. This invention combines the characteristics of multi-tenant microservice architecture, constructs a tensor structure representation of service call paths, introduces a path attention mechanism and a context gating modeling module, and employs the Spotted Hyena optimization algorithm to perform global search and multi-objective tuning of structural parameters. The invention details the structural optimization process for dynamic service path modeling, possessing advantages such as high modeling flexibility, high deployment efficiency, and strong adaptability to multiple scenarios.

[0007] The enterprise digitalization method based on SaaS microservice architecture according to embodiments of the present invention includes the following steps:

[0008] S1. Deploy a microservice system that supports multi-tenant access in an enterprise-level SaaS platform, collect the service call sequence of each tenant, and construct a path transition tensor representing the multi-hop call relationship between services;

[0009] S2. Using the path transition tensor as input, construct a path modeling network model that includes path attention mechanism and context gating mechanism. The structural parameters of the path modeling network model include jump weights, connection weights and gating thresholds.

[0010] S3. Initialize the spotted hyena population, encoding a set of structural parameters for each individual in the population;

[0011] S4. Apply the structural parameters of each hyena individual to the model, perform path prediction on the validation set, and evaluate the individual fitness based on prediction accuracy, path connectivity score and abnormal path identification rate.

[0012] S5. Based on the trapping and encirclement update mechanism of the spotted hyena optimization algorithm, and based on the structural parameters of the individual with the best fitness, guide the other individuals to perform difference transformation and perturbation update on the encoded jump weights, connection weights and gating thresholds.

[0013] S6. Repeat the model reconstruction, prediction evaluation and parameter update process until the iteration termination condition is met, extract the structural parameter set with the best fitness value, and deploy it to the service orchestration module of the SAAS platform to realize tenant-level path prediction and microservice process optimization.

[0014] Optionally, S1 specifically includes:

[0015] S11. In a microservice system deployed on an enterprise-level SaaS platform, service access log data generated by several tenant users during actual business processes is received. The log data includes timestamps, tenant identifiers, service call start points, and service call end points.

[0016] S12. Aggregate the service call logs of each tenant in chronological order to generate a service call sequence set. ,in Indicates the first The call path sequence of a tenant consists of a set of service nodes arranged in the order of the calls;

[0017] S13, Based on service call sequence set Construct a third-order path transition tensor T', which consists of elements composition, Indicates from service node To service node Passing through intermediate jump nodes The transition weights are obtained by normalizing the frequency of occurrence in all path sequences.

[0018] Optionally, S2 specifically includes:

[0019] S21. Transfer the third-order path transition tensor 'Input into the path modeling network as the basic representation structure of the jump relationship between service nodes;'

[0020] S22. Introduce a path attention mechanism into the path modeling network and define the service connection weight matrix. ,in Indicates from service node To service node Attention connection weights, the connection weight matrix With jump weight tensor Joint operation is used for multi-hop path information aggregation;

[0021] S23. Configure a context gating mechanism in the network structure and construct a gating threshold vector. ,in Indicates service node The corresponding context activation threshold is used to regulate the degree of activation of the tenant's behavioral context on the path state;

[0022] S24, Change the jump weight tensor Connection weight matrix With gate threshold vector Together as a set of structural parameters , used to represent the structural adjustable parameters of the path modeling network.

[0023] Optionally, S3 specifically includes:

[0024] S31, Based on structural parameter set Constructing parameter vector representation ,in Represents a complete structural parameter individual, with the parameter vector having dimension . Parameter vector dimension It is composed of jump weight tensor Connection weight matrix and gate threshold vector The total dimension obtained after flattening and stitching;

[0025] S32, Set the population size as follows , generate containing Individual spotted hyena population Each of them It is a set of structural parameters that serve as the input parameters to be optimized in the path modeling network;

[0026] S33. For each individual in the initial population The values ​​of each dimension are within the set range Uniform random sampling initialization is performed within the specified time. , These represent the lower and upper bounds of each structural parameter dimension, respectively, to ensure the validity of parameter generation and the trainability of the model.

[0027] Optionally, S4 specifically includes:

[0028] S41. Combining structural parameters In the input path modeling network model, constructing a network based on... The prediction model configured with parameters performs a multi-hop service path prediction task on the validation dataset, and outputs a set of predicted path sequences and the activation status of service nodes.

[0029] S42. Based on the model output, calculate the results for each individual. Corresponding path prediction accuracy Service path connectivity score and abnormal path identification rate These are used to measure path matching degree, structural continuity, and anomaly detection capability, respectively.

[0030] S43. Constructing the fitness function

[0031] ;

[0032] in, For the weighting coefficients, satisfying This is used to control the weighting of each performance indicator in the overall fitness assessment.

[0033] S44. Set the fitness value of each individual. This serves as a performance benchmark in the spotted hyena optimization algorithm for selecting optimal structural parameters and driving population evolution.

[0034] Optionally, S5 specifically includes:

[0035] S51, from the individual's fitness value Select the individual with the highest fitness value. Let the current optimal structure parameter individual be denoted as its structure parameter set. , set the structural parameters Flatten and splice into a structural parameter vector ;

[0036] S52. For each non-optimal individual The structural parameter update operation is performed, and the individual encoding vector is updated using the difference transformation strategy in the spotted hyena optimization algorithm. The update formula is as follows:

[0037] ;

[0038] in, This is the contraction vector, which controls the global convergence speed. This is a perturbation vector used to simulate the uncertain behavior during encirclement. For the first Individual parameters in round iteration;

[0039] S53, Update the individual Decoding into a combination of structural parameters Maintain consistency with the initial structure dimensions;

[0040] S54. Use the updated structural parameter set of all individuals as input again to execute the next round of path modeling network training and fitness evaluation tasks, as the basis for iteration.

[0041] Optionally, the difference transformation strategy in S52 specifically includes:

[0042] S521. Flatten and concatenate each individual to be updated into a structural parameter vector. Get the individual with the highest fitness value in the current population. And flatten and splice them into a structural parameter vector Construct the target difference vector ,,in The perturbation coefficient vector, denoted by " " indicates a multiplication operation in the corresponding dimension. For the first The perturbation factor follows a uniform distribution;

[0043] S522. Calculate the convergence compression amplitude vector. ,in The convergence factor vector, With the number of iterations Monotonically decreasing, controlling the first Degree of compression of dimensional parameters;

[0044] S523. Based on the optimal individual and the compression amplitude, update the current individual parameter vector as follows: ;Will As a new generation of structural parameter encoding, it is sent to the next round of model building and fitness evaluation process.

[0045] Optionally, S6 specifically includes:

[0046] S61. Set the maximum number of iteration rounds and the minimum fitness improvement threshold. As the termination condition of the spotted hyena optimization algorithm, fitness improvement is defined as the difference between the optimal fitness in two consecutive rounds. ;

[0047] S62. When any termination condition is met, the final optimal individual parameter vector is... Decoded into a set of structural parameters ,in, For the jump weight tensor of the final optimal individual parameter vector, The connection weight matrix is ​​the final optimal individual parameter vector. The gate threshold vector is the final optimal individual parameter vector;

[0048] S63, Set the structural parameter group The final deployment version of the path modeling network model is applied to the network model. After complete training, the network model weights are saved to generate a path prediction model and a service status evaluation model.

[0049] S64. Integrate the service status assessment model into the service orchestration subsystem of the SAAS platform, and run it as a dynamic path prediction module to support real-time prediction of service call paths and microservice node weight assessment tasks in a multi-tenant environment.

[0050] An enterprise digital system based on a SaaS microservice architecture according to an embodiment of the present invention includes the following modules:

[0051] The service data acquisition module is used to collect service call logs from multiple tenants and generate service path sequence data.

[0052] The path structure generation module is used to construct a structural representation of service jump relationships based on service path data.

[0053] The parameter modeling module is used to configure the structural parameters of the path modeling network, including jump relationships, connection weights, and gating thresholds.

[0054] The population initialization module is used to initialize an optimized population containing multiple combinations of structural parameters.

[0055] The performance evaluation module is used to perform path prediction tasks and calculate the performance metrics corresponding to the structural parameters.

[0056] The optimization and update module is used to guide the iterative update of other parameters based on the optimal parameters.

[0057] The convergence control module is used to determine whether the optimization process meets the requirements for the number of iterations or performance improvement.

[0058] The system deployment module is used to integrate the optimized path model into the SaaS platform for service process scheduling.

[0059] The beneficial effects of this invention are:

[0060] (1) By constructing a high-dimensional tensor structure of service call path and introducing path attention mechanism and context gating mechanism, this invention can effectively capture the jump relationship and potential dependency between microservices, realize fine modeling of multi-tenant service behavior, and break through the technical bottleneck of insufficient processing capability of traditional sequence model for complex path structure.

[0061] (2) The present invention uses the spotted hyena optimization algorithm to globally optimize the structural parameters in the path modeling network. By dynamically adjusting the connection weights and control thresholds through an evolutionary search strategy, it achieves a synergistic improvement of multiple performance indicators (such as prediction accuracy, path coherence and anomaly detection capability), overcoming the problems of slow convergence and easy trapping in local optima in traditional optimization methods.

[0062] (3) The modeling and optimization method proposed in this invention can be directly integrated into the service scheduling system of the SAAS platform. It has modular deployment and structural adaptability, and can be widely applied to multi-tenant service flow modeling and automatic orchestration in different industry scenarios, significantly improving the system's deployment efficiency, model stability and business adaptability. Attached Figure Description

[0063] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0064] Figure 1 This is an overall flowchart of the enterprise digitalization method based on SaaS microservice architecture proposed in this invention;

[0065] Figure 2 This is a schematic diagram of the path modeling network structure for the enterprise digitalization method based on SaaS microservice architecture proposed in this invention.

[0066] Figure 3 This is a flowchart illustrating the structural parameter optimization process based on the spotted hyena optimization algorithm for the enterprise digitalization method based on SaaS microservice architecture proposed in this invention. Detailed Implementation

[0067] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0068] refer to Figure 1-3 The enterprise digital transformation approach based on SaaS microservice architecture includes the following steps:

[0069] S1. Deploy a microservice system that supports multi-tenant access in an enterprise-level SaaS platform, collect the service call sequence of each tenant, and construct a path transition tensor representing the multi-hop call relationship between services;

[0070] S2. Using the path transition tensor as input, construct a path modeling network model that includes path attention mechanism and context gating mechanism. The structural parameters of the path modeling network model include jump weights, connection weights and gating thresholds.

[0071] S3. Initialize the spotted hyena population, encoding a set of structural parameters for each individual in the population;

[0072] S4. Apply the structural parameters of each hyena individual to the model, perform path prediction on the validation set, and evaluate the individual fitness based on prediction accuracy, path connectivity score and abnormal path identification rate.

[0073] S5. Based on the trapping and encirclement update mechanism of the spotted hyena optimization algorithm, and based on the structural parameters of the individual with the best fitness, guide the other individuals to perform difference transformation and perturbation update on the encoded jump weights, connection weights and gating thresholds.

[0074] S6. Repeat the model reconstruction, prediction evaluation and parameter update process until the iteration termination condition is met, extract the structural parameter set with the best fitness value, and deploy it to the service orchestration module of the SAAS platform to realize tenant-level path prediction and microservice process optimization.

[0075] This invention constructs a tensor structure representation centered on jump paths and combines a context gating mechanism with the Spotted Hyena optimization algorithm to achieve structural adaptation and parameter optimization in microservice path modeling. Compared to traditional static sequence-based modeling methods, this invention overcomes the problems of rigid service call modeling and insufficient optimization accuracy. It can flexibly adjust the model structure in multi-tenant SaaS scenarios, improve the accuracy of service path analysis and system deployment efficiency, and has significant versatility and scalability.

[0076] In this embodiment, S1 specifically includes:

[0077] S11. In a microservice system deployed on an enterprise-level SaaS platform, service access log data generated by several tenant users during actual business processes is received. The log data includes timestamps, tenant identifiers, service call start points, and service call end points.

[0078] S12. Aggregate the service call logs of each tenant in chronological order to generate a service call sequence set. ,in Indicates the first The call path sequence of a tenant consists of a set of service nodes arranged in the order of the calls;

[0079] S13, Based on service call sequence set Construct a third-order path transition tensor T', which consists of elements composition, Indicates from service node To service node Passing through intermediate jump nodes The transition weights are obtained by normalizing the frequency of occurrence in all path sequences.

[0080] This invention specifically considers the diversity of tenant granularity and service configuration dimensions during the service call data acquisition process, mapping service call sequences and configuration parameters to a unified tensor input structure to ensure consistent adaptability of the modeling process across different business scenarios. Compared to existing technologies, this invention introduces a context enhancement mechanism in the raw data preprocessing stage, making the subsequent model's understanding of service behavior more accurate and significantly enhancing the robustness of the modeling.

[0081] In this embodiment, S2 specifically includes:

[0082] S21. Transfer the third-order path transition tensor 'Input into the path modeling network as the basic representation structure of the jump relationship between service nodes;'

[0083] S22. Introduce a path attention mechanism into the path modeling network and define the service connection weight matrix. ,in Indicates from service node To service node Attention connection weights, the connection weight matrix With jump weight tensor Joint operation is used for multi-hop path information aggregation;

[0084] S23. Configure a context gating mechanism in the network structure and construct a gating threshold vector. ,in Indicates service node The corresponding context activation threshold is used to regulate the degree of activation of the tenant's behavioral context on the path state;

[0085] S24, Change the jump weight tensor Connection weight matrix With gate threshold vector Together as a set of structural parameters , used to represent the structural adjustable parameters of the path modeling network.

[0086] This invention constructs a multi-layered network comprising a jump modeling layer, a path attention mechanism, and a context gating structure to perform multi-dimensional feature extraction and semantic modeling of service paths. Compared to ordinary graph neural networks or sequence models, this structure is more suited to the characteristics of discontinuous calls in microservices, achieving a better balance between structural expressiveness and inference efficiency, and significantly improving the accuracy and interpretability of path recognition.

[0087] In this embodiment, S3 specifically includes:

[0088] S31, Based on structural parameter set Constructing parameter vector representation ,in Represents a complete structural parameter individual, with the parameter vector having dimension . Parameter vector dimension It is composed of jump weight tensor Connection weight matrix and gate threshold vector The total dimension obtained after flattening and stitching;

[0089] S32, Set the population size as follows , generate containing Individual spotted hyena population Each of them It is a set of structural parameters that serve as the input parameters to be optimized in the path modeling network;

[0090] S33. For each individual in the initial population The values ​​of each dimension are within the set range Uniform random sampling initialization is performed within the specified time. , These represent the lower and upper bounds of each structural parameter dimension, respectively, to ensure the validity of parameter generation and the trainability of the model.

[0091] This invention introduces a multi-objective structural optimization function, jointly considering accuracy, path coherence, and anomaly detection capabilities, providing a quantitative basis for model structure design. Unlike traditional methods that only use loss functions as evaluation criteria, this scheme can flexibly adjust weights to achieve adaptive trade-offs under multiple objectives, significantly improving the model's engineering adaptability and algorithm stability, and meeting the deployment needs of SaaS in multiple scenarios.

[0092] In this embodiment, S4 specifically includes:

[0093] S41. Combining structural parameters In the input path modeling network model, constructing a network based on... The prediction model configured with parameters performs a multi-hop service path prediction task on the validation dataset, and outputs a set of predicted path sequences and the activation status of service nodes.

[0094] S42. Based on the model output, calculate the results for each individual. Corresponding path prediction accuracy Service path connectivity score and abnormal path identification rate These are used to measure path matching degree, structural continuity, and anomaly detection capability, respectively.

[0095] S43. Constructing the fitness function

[0096] ;

[0097] in, For the weighting coefficients, satisfying This is used to control the weighting of each performance indicator in the overall fitness assessment.

[0098] S44. Set the fitness value of each individual. This serves as a performance benchmark in the spotted hyena optimization algorithm for selecting optimal structural parameters and driving population evolution.

[0099] This invention employs the Spotted Hyena optimization algorithm to search and optimize structural parameters, combined with multi-objective performance feedback of the service path tensor, to achieve global optimization and dynamic adjustment of structural parameters. This algorithm exhibits high global convergence and strong solution space jump capability, outperforming traditional genetic algorithms or particle swarm optimization algorithms. It effectively addresses the problems of insufficient structural search and optimization getting trapped in local optima in microservice path modeling.

[0100] In this embodiment, S5 specifically includes:

[0101] S51, from the individual's fitness value Select the individual with the highest fitness value. Let the current optimal structure parameter individual be denoted as its structure parameter set. , set the structural parameters Flatten and splice into a structural parameter vector ;

[0102] S52. For each non-optimal individual The structural parameter update operation is performed, and the individual encoding vector is updated using the difference transformation strategy in the spotted hyena optimization algorithm. The update formula is as follows:

[0103] ;

[0104] in, This is the contraction vector, which controls the global convergence speed. This is a perturbation vector used to simulate the uncertain behavior during encirclement. For the first Individual parameters in round iteration;

[0105] S53, Update the individual Decoding into a combination of structural parameters Maintain consistency with the initial structure dimensions;

[0106] S54. Use the updated structural parameter set of all individuals as input again to execute the next round of path modeling network training and fitness evaluation tasks, as the basis for iteration.

[0107] This invention introduces a precision difference transformation mechanism during the model training phase, guiding the training objective towards the structural adjustment process, thereby improving the model's modeling ability under boundary samples. This strategy effectively alleviates the overfitting problem during training, while enhancing the model's generalization ability to differences in tenant behavior, improving the system's cross-tenant modeling performance and service consistency.

[0108] In this embodiment, the difference transformation strategy in S52 specifically includes:

[0109] S521. Flatten and concatenate each individual to be updated into a structural parameter vector. Get the individual with the highest fitness value in the current population. And flatten and splice them into a structural parameter vector Construct the target difference vector ,,in The perturbation coefficient vector, denoted by " " indicates a multiplication operation in the corresponding dimension. For the first The perturbation factor follows a uniform distribution;

[0110] S522. Calculate the convergence compression amplitude vector. ,in The convergence factor vector, With the number of iterations Monotonically decreasing, controlling the first Degree of compression of dimensional parameters;

[0111] S523. Based on the optimal individual and the compression amplitude, update the current individual parameter vector as follows: ;Will As a new generation of structural parameter encoding, it is sent to the next round of model building and fitness evaluation process.

[0112] In this embodiment, S6 specifically includes:

[0113] S61. Set the maximum number of iteration rounds and the minimum fitness improvement threshold. As the termination condition of the spotted hyena optimization algorithm, fitness improvement is defined as the difference between the optimal fitness in two consecutive rounds. ;

[0114] S62. When any termination condition is met, the final optimal individual parameter vector is... Decoded into a set of structural parameters ,in, For the jump weight tensor of the final optimal individual parameter vector, The connection weight matrix is ​​the final optimal individual parameter vector. The gate threshold vector is the final optimal individual parameter vector;

[0115] S63, Set the structural parameter group The final deployment version of the path modeling network model is applied to the network model. After complete training, the network model weights are saved to generate a path prediction model and a service status evaluation model.

[0116] S64. Integrate the service status assessment model into the service orchestration subsystem of the SAAS platform, and run it as a dynamic path prediction module to support real-time prediction of service call paths and microservice node weight assessment tasks in a multi-tenant environment.

[0117] The service scheduling module constructed in this invention combines modeling results to achieve deployment mapping of path optimization results, forming a closed-loop response mechanism between structural parameters and microservice scheduling logic, thus ensuring the coupling efficiency between the model and the platform. Unlike traditional model deployments that require extensive manual configuration, this invention enables automatic mapping from paths to microservices, effectively improving deployment speed and scheduling accuracy.

[0118] An enterprise digital system based on a SaaS microservice architecture according to an embodiment of the present invention includes the following modules:

[0119] The service data acquisition module is used to collect service call logs from multiple tenants and generate service path sequence data.

[0120] The path structure generation module is used to construct a structural representation of service jump relationships based on service path data.

[0121] The parameter modeling module is used to configure the structural parameters of the path modeling network, including jump relationships, connection weights, and gating thresholds.

[0122] The population initialization module is used to initialize an optimized population containing multiple combinations of structural parameters.

[0123] The performance evaluation module is used to perform path prediction tasks and calculate the performance metrics corresponding to the structural parameters.

[0124] The optimization and update module is used to guide the iterative update of other parameters based on the optimal parameters.

[0125] The convergence control module is used to determine whether the optimization process meets the requirements for the number of iterations or performance improvement.

[0126] The system deployment module is used to integrate the optimized path model into the SaaS platform for service process scheduling.

[0127] The system structure of this invention is complete, with clearly defined module responsibilities, and features a closed-loop architecture from data acquisition to path optimization and deployment execution. This design not only improves the system's functional integration but also enhances the platform's stability and scalability when facing multi-tenant concurrency and business differences.

[0128] Example 1:

[0129] To verify the feasibility of this invention in practice, it was applied to a digital management platform transformation project of a manufacturing enterprise. The enterprise's original information system consisted of multiple monolithic applications, with problems such as data silos, redundant service deployments, and low response efficiency among the various business subsystems, severely restricting the enterprise's operational efficiency and system maintainability. To address this situation, the "Enterprise Digital System and Method Based on SaaS Microservice Architecture" proposed in this invention was used for its reconstruction and deployment.

[0130] In this scenario, service call logs from multiple business domains, including production, procurement, warehousing, and sales, are first collected. Multi-hop service paths are then extracted and a service path tensor structure is generated. Based on this constructed jump path tensor structure, the path modeling network proposed in this invention is applied to deeply model the contextual dependencies and importance of service paths. After the model structure is determined, the Spotted Hyena optimization algorithm is used to perform multi-objective global optimization of key parameters of the modeling network (such as jump window width, number of attention heads, and gating coefficients), achieving a comprehensive optimization of the model in terms of accuracy, stability, and response speed.

[0131] The optimized model was deployed on the enterprise's SaaS platform. The system supports multi-tenant access and can adaptively generate service paths and structural parameter mappings based on tenant behavior. Within three weeks of the platform's launch, the average system service response latency decreased from 780ms to 415ms, and the average throughput increased by 1.6 times. The F1-score for path modeling improved from 0.77 to 0.89, achieving an accuracy of 93.6% in abnormal path detection tasks. The success rate of microservice path identification increased to 97.4%, and the overall system downtime decreased by 42%.

[0132] To further analyze the practical effects of this invention, the following table provides a comparison of data before and after deployment in different modules, covering three dimensions: performance, stability, and intelligence level, demonstrating the technical superiority of this invention in actual production scenarios.

[0133] Table 1: Comparison of System Performance and Structure Optimization Effects Before and After Enterprise-Level SaaS Platform Deployment

[0134] Indicator Name Existing system values numerical values ​​after applying this invention Improvement rate / magnitude of change Average service response time (ms) 780 415 ↓46.8% Single-tenant service call throughput (times / s) 112 179 ↑59.8% Path modeling accuracy (F1-score) 0.77 0.89 ↑15.6% Accuracy rate of abnormal path identification (%) 84.2% 93.6% ↑11.2% Model structure iteration count (rounds) 35 22 ↓37.1% Path recognition success rate (%) 88.5% 97.4% ↑10.1% Model structure parameter dimension 960 684 ↓28.8% Structural optimization time (seconds) 126 83 ↓34.1% Average annual system downtime (hours) 14.5 8.4 ↓42.1% Multi-tenant service separation score (0-1) 0.61 0.87 ↑42.6%

[0135] In terms of overall service response efficiency, after adopting the method of this invention, the average service response latency of the system was reduced from 780 milliseconds to 415 milliseconds, a reduction of 46.8%. This significant optimization is due to the efficient modeling of service dependency paths by the jump path tensor structure, which can identify high-frequency call patterns in advance and respond quickly in the model, avoiding redundant calculations and inefficient path delays.

[0136] In terms of system throughput, the single-tenant service call throughput increased from 112 calls per second to 179 calls per second, an increase of nearly 60%. This improvement reflects the enhanced ability of the system to handle concurrent service requests after structural optimization. The call process after path modeling is simpler, resource scheduling is more reasonable, and the performance bottleneck of microservice systems in high-concurrency scenarios is effectively alleviated.

[0137] This invention also demonstrates significant advantages in path recognition accuracy. By introducing context gating and path attention mechanisms, the modeling accuracy is further improved, with the F1-score increasing from 0.77 to 0.89. Simultaneously, in the abnormal path recognition task, the accuracy reaches 93.6%, an improvement of approximately 11 percentage points compared to the traditional solution's 84.2%, indicating that the optimized structure can better capture potential abnormal patterns in the service link.

[0138] From a model training perspective, the optimized model reduced the number of structural iterations from 35 to 22, shortening the training cycle by 37.1%. Simultaneously, the model's structural parameter dimension decreased from 960 to 684, reducing model complexity by 28.8%. This improvement not only accelerated deployment but also reduced system load and improved model inference efficiency.

[0139] Furthermore, the time taken after structural optimization was reduced from 126 seconds to 83 seconds, and the downtime was reduced from an average of 14.5 hours per year to 8.4 hours, a reduction of more than 40%. This indicates that the present invention enhances system stability and fault tolerance while ensuring high availability.

[0140] Finally, the platform's multi-tenant service separation score improved from 0.61 to 0.87, reflecting that the microservice path model constructed in this invention can more clearly distinguish and manage service flow logic in a multi-tenant environment, significantly improve the problems of interference and resource isolation between tenants, and provide reliable support for the large-scale deployment of enterprise-level platforms.

[0141] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An enterprise digitalization approach based on a SaaS microservice architecture, characterized in that: Includes the following steps: S1. Deploy a microservice system that supports multi-tenant access in an enterprise-level SaaS platform, collect the service call sequence of each tenant, and construct a path transition tensor representing the multi-hop call relationship between services; S2. Using the path transition tensor as input, construct a path modeling network model that includes path attention mechanism and context gating mechanism. The structural parameters of the path modeling network model include jump weights, connection weights and gating thresholds. S3. Initialize the spotted hyena population, encoding a set of structural parameters for each individual in the population; S4. Apply the structural parameters of each hyena individual to the model, perform path prediction on the validation set, and evaluate the individual fitness based on prediction accuracy, path connectivity score and abnormal path identification rate. S5. Based on the trapping and encirclement update mechanism of the spotted hyena optimization algorithm, and based on the structural parameters of the individual with the best fitness, guide the other individuals to perform difference transformation and perturbation update on the encoded jump weights, connection weights and gating thresholds. S6. Repeat the model reconstruction, prediction evaluation and parameter update process until the iteration termination condition is met, extract the structural parameter set with the best fitness value, and deploy it to the service orchestration module of the SAAS platform to realize tenant-level path prediction and microservice process optimization.

2. The enterprise digitalization method based on SaaS microservice architecture according to claim 1, characterized in that, S1 specifically includes: S11. In a microservice system deployed on an enterprise-level SaaS platform, service access log data generated by several tenant users during actual business processes is received. The log data includes timestamps, tenant identifiers, service call start points, and service call end points. S12. Aggregate the service call logs of each tenant in chronological order to generate a service call sequence set. ,in Indicates the first The call path sequence of a tenant consists of a set of service nodes arranged in the order of the calls; S13, Based on service call sequence set Construct a third-order path transition tensor T', which consists of elements composition, Indicates from service node To service node Passing through intermediate jump nodes The transition weights are obtained by normalizing the frequency of occurrence in all path sequences.

3. The enterprise digitalization method based on SaaS microservice architecture according to claim 1, characterized in that, S2 specifically includes: S21. Transfer the third-order path transition tensor 'Input into the path modeling network as the basic representation structure of the jump relationship between service nodes;' S22. Introduce a path attention mechanism into the path modeling network and define the service connection weight matrix. ,in Indicates from service node To service node Attention connection weights, the connection weight matrix With jump weight tensor Joint operation is used for multi-hop path information aggregation; S23. Configure a context gating mechanism in the network structure and construct a gating threshold vector. ,in Indicates service node The corresponding context activation threshold is used to regulate the degree of activation of the tenant's behavioral context on the path state; S24, Change the jump weight tensor Connection weight matrix With gate threshold vector Together as a set of structural parameters , used to represent the structural adjustable parameters of the path modeling network.

4. The enterprise digitalization method based on SaaS microservice architecture according to claim 1, characterized in that, S3 specifically includes: S31, Based on structural parameter set Constructing parameter vector representation ,in Represents a complete structural parameter individual, where, For the jump weight tensor, To connect the weight matrix, This is the gate threshold vector; S32, Set the population size as follows , generate containing Individual spotted hyena population Each of them It is a set of structural parameters that serve as the input parameters to be optimized in the path modeling network; S33. For each individual in the initial population The values ​​of each dimension are within the set range Uniform random sampling initialization is performed within the specified time. , These represent the lower and upper bounds of each structural parameter dimension, respectively, to ensure the validity of parameter generation and the trainability of the model.

5. The enterprise digitalization method based on SaaS microservice architecture according to claim 1, characterized in that, S4 specifically includes: S41. Combining structural parameters In the input path modeling network model, constructing a network based on... The prediction model configured with parameters performs a multi-hop service path prediction task on the validation dataset, and outputs a set of predicted path sequences and the activation status of service nodes. S42. Based on the model output, calculate the results for each individual. Corresponding path prediction accuracy Service path connectivity score and abnormal path identification rate These are used to measure path matching degree, structural continuity, and anomaly detection capability, respectively. S43. Constructing the fitness function ； in, For the weighting coefficients, satisfying This is used to control the weighting of each performance indicator in the overall fitness assessment. S44. Set the fitness value of each individual. This serves as a performance benchmark in the spotted hyena optimization algorithm for selecting optimal structural parameters and driving population evolution.

6. The enterprise digitalization method based on SaaS microservice architecture according to claim 5, characterized in that, S5 specifically includes: S51, from the individual's fitness value Select the individual with the highest fitness value. Let the current optimal structure parameter individual be denoted as its structure parameter set. , set the structural parameters Flatten and splice into a structural parameter vector ;in, For the jump weight tensor of the individual with optimal structure parameters The connection weight matrix for the individual with optimal structural parameters. For each individual with optimal structural parameters, the gate threshold vector is used. S52. For each non-optimal individual The structural parameter update operation is performed, and the individual encoding vector is updated using the difference transformation strategy in the spotted hyena optimization algorithm. The update formula is as follows: ； in, This is the contraction vector, which controls the global convergence speed. This is a perturbation vector used to simulate the uncertain behavior during encirclement. For the first Individual parameters in round iteration; S53, Update the individual Decoding into a combination of structural parameters Maintain consistency with the initial structure dimensions; among which, Let i be the set of structure parameters updated for the i-th individual in the (t+1)-th iteration. Let i be the jump weight tensor of the i-th individual after the (t+1)-th iteration. Let be the connection weight matrix of the i-th individual after the (t+1)-th iteration. Let be the updated gating threshold vector for the i-th individual in the (t+1)-th iteration; S54. Use the updated structural parameter set of all individuals as input again to execute the next round of path modeling network training and fitness evaluation tasks, as the basis for iteration.

7. The enterprise digitalization method based on SaaS microservice architecture according to claim 6, characterized in that, The difference transformation strategy in S52 specifically includes: S521. Flatten and concatenate each individual to be updated into a structural parameter vector. Get the individual with the highest fitness value in the current population. And flatten and splice them into a structural parameter vector Construct the target difference vector ,,in The vector of perturbation coefficients, denoted by " " indicates a multiplication operation in the corresponding dimension. For the first The perturbation factor follows a uniform distribution; S522. Calculate the convergence compression amplitude vector. ,in The convergence factor vector, With the number of iterations Monotonically decreasing, controlling the first Degree of compression of dimensional parameters; S523. Based on the optimal individual and the compression amplitude, update the current individual parameter vector as follows: ;Will As a new generation of structural parameter encoding, it is sent to the next round of model building and fitness evaluation process.

8. The enterprise digitalization method based on SaaS microservice architecture according to claim 1, characterized in that, S6 specifically includes: S61. Set the maximum number of iteration rounds and the minimum fitness improvement threshold. As the termination condition of the spotted hyena optimization algorithm, fitness improvement is defined as the difference between the optimal fitness in two consecutive rounds. ; S62. When any termination condition is met, the final optimal individual parameter vector is... Decoded into a set of structural parameters ,in, For the jump weight tensor of the final optimal individual parameter vector, The connection weight matrix is ​​the final optimal individual parameter vector. The gate threshold vector is the final optimal individual parameter vector; S63, Set the structural parameter group The final deployment version of the path modeling network model is applied to the network model. After complete training, the network model weights are saved to generate a path prediction model and a service status evaluation model. S64. Integrate the service status assessment model into the service orchestration subsystem of the SAAS platform, and run it as a dynamic path prediction module to support real-time prediction of service call paths and microservice node weight assessment tasks in a multi-tenant environment.

9. An enterprise digital system based on a SaaS microservice architecture, applied to any one of the enterprise digital methods based on a SaaS microservice architecture as described in claims 1 to 8, characterized in that, Includes the following modules: The service data acquisition module is used to collect service call logs from multiple tenants and generate service path sequence data. The path structure generation module is used to construct a structural representation of service jump relationships based on service path data. The parameter modeling module is used to configure the structural parameters of the path modeling network, including jump relationships, connection weights, and gating thresholds. The population initialization module is used to initialize an optimized population containing multiple combinations of structural parameters. The performance evaluation module is used to perform path prediction tasks and calculate the performance metrics corresponding to the structural parameters. The optimization and update module is used to guide the iterative update of other parameters based on the optimal parameters. The convergence control module is used to determine whether the optimization process meets the requirements for the number of iterations or performance improvement. The system deployment module is used to integrate the optimized path model into the SaaS platform for service process scheduling.

Images