Improved RNN exchange cloud computing power network efficient semantic interoperation method

Abstract

The invention relates to an efficient interoperability research method and system for a semantic ontology in a cloud environment based on a recurrent neural network of an improved algorithm. The method comprises the following steps: standardizing data of an application and a cloud server stored in a data warehouse; establishing a semantic layer to provide unified interpretation for the data; the optimal feature is selected by using a GCM-IFF algorithm; constructing an application dynamic ontology and generating an OWL file; ontology operation and reasoning are completed by using a HermiT inference device; the cloud environment semantic interoperability training prediction exchange optimization algorithm is used for optimization, the RNN gradient problem can be effectively solved, the method has the advantages of being high in precision (the accuracy rate is 96.32%), low in calculation amount (the training time is 22567 ms), low in memory usage amount (17515 kb), high in robustness and the like, and the working efficiency of the cloud environment semantic interoperability is improved.

Description

Technical Field

[0001] This invention relates to the field of cloud computing (CC) technology, specifically to a method and system for achieving semantic interoperability (SI) between applications and cloud servers (CS) in a cloud environment. More particularly, it relates to a technique that combines an improved exchange optimization algorithm with a recurrent neural network (RNN) to achieve efficient and accurate semantic interoperability by constructing an ontology. Background Technology

[0002] Cloud computing is a crucial technology encompassing many aspects, such as container technology, serverless technology, microservices, IoT, edge computing, and artificial intelligence. The widespread adoption of cloud services requires certain prerequisites: seamless integration and data exchange between different cloud service providers—a key aspect of interoperability. Furthermore, interoperability is essential for devices and platforms in cloud manufacturing and other scenarios to collaborate effectively and maximize the advantages of the cloud environment.

[0003] In existing technologies, achieving semantic interoperability in cloud environments largely relies on standardization efforts, common data formats, communication protocols, and so on. However, these methods have significant drawbacks:

[0004] 1. Lack of flexibility and rigid standardized structures make it difficult to meet the diverse and frequently changing needs of cloud applications, hindering the application of new skills and the expansion of new cloud environments.

[0005] 2. Inadequate semantic capture: Traditional methods struggle to capture the semantics of the vast amounts of data and services in the cloud environment. The cloud ecosystem contains structured databases and various multimedia materials of different types, making it difficult to handle the subtle relationships and semantics using existing methods, resulting in suboptimal interoperability.

[0006] 3. Poor feature selection efficiency: Existing methods struggle to efficiently and accurately select relevant features in cloud environments, which reduces the accuracy of interoperability and increases computation time.

[0007] 4. The creation and manipulation of ontology are not standardized. There are no standard methods for creating and manipulating ontology, resulting in inconsistent data representation and affecting the accuracy and consistency of semantic interoperability.

[0008] 5. Computational efficiency and storage issues: Some existing methods have high storage requirements, are complex to implement, and have limited applications.

[0009] Therefore, there is an urgent need for a new approach that can effectively address the problem and improve the accuracy, speed, and scalability of semantic interoperability in cloud environments. Summary of the Invention

[0010] The purpose of this invention is to provide a cloud-based method and system for efficient ontology semantic interoperability based on an improved recurrent neural network for shell games, thereby solving the problems of insufficient semantic interoperability accuracy, slow computation speed, and lack of standardization in ontology creation in current technologies.

[0011] To achieve the above objectives, this invention provides the following technical solution:

[0012] A method for semantic interoperability in cloud environments using an improved exchange optimization algorithm recurrent neural network includes the following steps:

[0013] 1. Data Warehouse Standardization: Data from applications and cloud servers is acquired from the data warehouse, standardized, and its quality is improved through data extraction, data cleaning, data transformation, data loading, and data refreshing. This process can be described as follows:

[0014] γ_DWS(list(Γ_n))=γ_DE(list(Γ_n))+γ_DC(list(Γ_n))+γ_DT(list(Γ_n))+γ_DL(list(Γ_n))+γ_REF(list(Γ_n))

[0015] Where γDWS is the standard data warehouse process, γDE is data extraction, γDC is data cleaning, γDT is data transformation, γDL is data loading, γREF is data updating, and list(Γn) is the data list.

[0016] 2. Establish a semantic layer: Put standardized data into the semantic layer (SL), which provides a unified method for interpreting data, establishes relationships between data, and creates a simple, business-oriented data view.

[0017] 3. Optimized Data Based on GCM-IFF: The Fertilizer Field Optimization Algorithm Based on Gene Mating Variants (GCM-IFF) selects the best features from the semantic layer data. The GCM-IFF optimization algorithm incorporates Gene Mating Variants (GCM) technology into the traditional Fertilizer Field Optimization (FF) algorithm to address the problem of FFO (Free Forward) easily getting trapped in local optima. Key data points are selected from the semantic layer data, such as data size, task cost, waiting time, turnaround time, cloud server memory, disk space, and the number of user requests. The objective function is:

[0018] Θ_objective=I_1Γ_size^D+I_2Γ_TC^D+I_3Γ_wt^D+I_4Γ_tat^D+I_5Γ_mem^c+I_6Γ_DS^c+I_7Γ_NR^c

[0019] Where Θ_objective is the objective function value, I_1-I_7 are weight constants (0-1), Γsize^D, ΓTC^D, Γwt^D, and Γtat^D represent the data size, total data cost, waiting time, and turnaround time, respectively, and Γmem^c, ΓDS^c, and ΓNR^c represent the cloud server's memory, disk, and number of user requests, respectively.

[0020] 4. Dynamic Application Ontology Construction: Based on the optimal features selected in step 3, the process of constructing a dynamic application ontology includes: determining the relationships between features; constructing a feature hierarchy based on the "is-a" attribute, mainly including the definition and differentiation of domain class attributes (object attributes and data attributes), the study of relationship details by refining logical expressions based on axioms, representing cloud instances, and consistency checking methods based on inference engines such as HermiT.

[0021] 5. OWL file generation: Use the protege tool to convert the constructed ontology structure into an OWL (Ontology Web Language) file so that machines can understand and process it.

[0022] 6. Cloud server interoperability and ontology operations based on HermiT: Ontology operations are performed using the HermiTOWL inferencer, including: reading OWL files; checking the consistency of classes and their attributes; converting OWL ontology into machine-understandable descriptive logical clauses; and reasoning about ontology consistency and inter-class inclusion relationships based on hypertable calculus.

[0023] 7. Cloud Interoperability Training and Prediction: Using the class features processed in step 6 as input, and an improved exchange optimization algorithm (Shell Game Optimization, an improved RNN) is input for training and testing to predict cloud server interoperability and select the optimal cloud server. Shell Game Optimization uses the exchange optimization algorithm (G_O) to select the "shell," i.e., the candidate weights, choosing the best one. Other selections are random. The accuracy of the player (weight) (AI_i(t)) is determined by their fitness value and probability of winning. ).

[0024] The weighted position is updated based on the rule of guessing the ball:

[0025]

[0026] Where r_1 and r_2 are random numbers between [0, 1]. This is based on the displacement of different shells.

[0027] This optimization effectively avoids the vanishing and exploding gradient problems of traditional RNNs. It does not require too many parameters to accelerate convergence and improve prediction accuracy. During training, the ReLU function and backpropagation are used to minimize the loss function.

[0028] Preferably, the data warehouse relies on the Apache Hadoop framework and uses the MapReduce processing model to achieve large-scale distributed processing of data.

[0029] As a preferred option, the semantic layer acts as a bridge between the data warehouse and reporting tools, simplifying self-service reporting, providing 80% data support, and supporting the remaining 20% ​​of data needs.

[0030] Alternatively, genetic crossover (P_C) and mutation (P_m) are introduced during the population update process to improve global search capabilities and prevent premature convergence. The crossover operation uses different formulas depending on the sign of the wind dispersion fraction (W_D).

[0031] The ontology was constructed using the preferred Protege tool, in which the ontology was added, and its built-in Hermit inference engine was used for ontology consistency and inference.

[0032] As the preferred input layer, class features processed by HermiT Hidden layers and output layers:

[0033]

[0034] Where χ_act represents the activation function (such as ReLU), For the bias term, w_ij(t) represents the weight.

[0035] Among the above technical solutions, this invention provides a cloud environment semantic interoperability method and system based on an improved exchange optimization algorithm recurrent neural network, which has the following technical advantages:

[0036] 1. High-precision interoperability: By constructing an ontology structure based on the semantic relationship between applications and cloud servers, and using it for accurate prediction, a cloud server selection accuracy of up to 96.32% was achieved (a significant improvement compared to existing methods such as DNN, CNN, RNN, and I-ANFIS), while obtaining 94.65% precision, 95.56% recall, 94.78% F-measure value, 95.32% sensitivity, and 96.22% specificity.

[0037] 2. Low computational overhead: The GCM-IFF algorithm effectively selects relevant features, reducing unnecessary dimensionality curse and computational burden caused by data; the exchange optimization algorithm optimizes RNN weights with fewer parameters and higher efficiency. Experiments show that the entire framework's training time is 22567ms, compared to existing methods such as DNN (26789ms), CNN (25678ms), RNN (24567ms), and I-ANFIS (23457ms).

[0038] 3. Low memory consumption: The optimized framework has a memory consumption of only 17,515 kb, which is much lower than other existing methods. DNN: 26,354 kb, CNN: 24,568 kb, RNN: 21,457 kb, I-ANFIS: 19,634 kb.

[0039] 4. Standardized Ontology Creation: The mature Protege tool and HermiTOWL inferencer are used for ontology creation and operation, and the consistency of the ontology is checked, thereby achieving standardization, reliability, and accurate and consistent semantic interoperability of ontology creation.

[0040] 5. Strong robustness: K-fold cross-validation analysis (1-5 fold accuracy rates of 97.99%, 97.96%, 97.93%, 96.95%, and 96.79%, respectively) showed that the present invention has stable performance in different data subsets and its robustness is higher than that of existing models (approximately 92.5%).

[0041] 6. Efficient Feature Selection: The GCM-IFF algorithm performs exceptionally well during iterative optimization. In the same number of iterations, it achieves more fitness values ​​than existing optimization methods such as Owl Search (OSA), Boolean Optimization (BWO), Tunica Algorithm (TSA), and Generalized Optimization Algorithm (GWFA). For example, after 25 iterations, the fitness value is 102 (OSA: 63, BWO: 73, TSA: 86, GWFA: 96).

[0042] 7. Effective solution to gradient problems: The exchange optimization algorithm effectively solves the problems of vanishing and exploding gradients in traditional RNNs, ensuring training stability and model performance. Attached Figure Description

[0043] The invention will now be further described with reference to the accompanying drawings.

[0044] Figure 1 This is a diagram showing the overall architecture of the system of the present invention;

[0045] Figure 2 This is a flowchart of the self-service analysis process for the semantic layer module of this invention;

[0046] Figure 3 This is a schematic diagram of the structure of the body construction module of the present invention;

[0047] Figure 4 This is an example diagram of the class hierarchy in the OWL ontology file of this invention.

[0048] Figure 5 The present invention comprises a data warehouse module 101 and a semantic layer module 102. Figure 6 This invention relates to a 105 cloud server interoperability module. Figure 7 This invention comprises a data standardization unit (201) and a business terminology mapping unit (202). Figure 8 The 103 feature selection module, 301 genetic crossover mutation unit, and 302 fertile field optimization unit of this invention are: Figure 9 This is a global module diagram of the present invention, illustrating the connections and data flow between various modules in the system, including the application service module, SQL database module, data warehouse module, semantic layer module, feature selection module, ontology construction module, ontology operation module, and cloud interoperability training and prediction module. 104 Ontology Construction Module, 401 Protege Editor, 402 HermiT Inferencer.

[0049] Figure 10This is a schematic diagram of the cloud interoperability training and prediction module in this invention, illustrating the process of weight optimization using an improved RNN model combined with a swap optimization algorithm, including the structure of the input layer, hidden layer, output layer, and weight update mechanism. Detailed Implementation

[0054] To facilitate those skilled in the art to understand the technical content of this invention patent, a detailed description of this invention patent will now be provided. This part is a further detailed description of the above technical solutions and an illustrative example of the invention content.

[0055] Example 1

[0056] This invention is implemented in a Java environment, using an Intelli5 / Core i7 processor with a CPU speed of 3.20 GHz, and running Windows 7. Of course, those skilled in the art will understand that the specific language, hardware, and operating system used to implement this invention can be selected as appropriate.

[0057] 1. System Architecture (corresponding) Figure 1 ):

[0058] App Service: A fully managed platform with built-in infrastructure, security patches, and extensions. Create code or containers using any language framework, leverage deployment slots to reduce downtime, and integrate mobile apps, websites, mobile services, and Biz-Talk services.

[0059] SQL database: Stores all detailed information about mobile applications and cloud servers, provides efficient query processing, is portable, and has multiple data overview formats. Use it to update the database or retrieve information according to mobile applications and cloud servers.

[0060] Data Warehouse (DW): Processes all information from all applications and cloud servers. It uses columnar storage with relational tables to reduce storage space and improve query performance. It's used to aggregate and store decision-making data generated by different organizations, processing large amounts of data from various sources. This implementation uses Apache Hadoop, which processes big data through its distributed data collection model (MapReduce). The MapReduce processing steps are as follows:

[0061] (i) The Map function (ν_map): takes data pairs in the input domain k1 as input and outputs a list of pairs in different domains k2: ν_map(k1, Γ1) = list(k2, Γ2).

[0062] (ii) The MapReduce framework collects and sorts all key-value pairs with the same key (k2).

[0063] (iii) Reduce function (ν_reduce): processes each group in parallel and generates a set of values ​​in the same domain: ν_reduce(k2,list(Γ2))→list(k3,Γ3).

[0064] Data Warehouse Standardization: Standardizing DW data to improve its quality. Perform in sequence:

[0065] (1) Data extraction (γ_DE): Data collection from heterogeneous sources

[0066] (2) Data cleaning (γ_DC): Detecting and correcting data errors

[0067] (3) Data Conversion (γ_DT): The process of converting data from legacy format to repository format.

[0068] (4) Data loading (γ_DL): Implement data grouping, summarizing, integrating, integrity checking, index building and partitioning.

[0069] (5) Data refresh (γ_REF): Update the data source to the repository.

[0070] The process is expressed as: γ_DWS(list(Γ_n))=γ_DE(list(Γ_n))+γ_DC(list(Γ_n))+γ_DT(list(Γ_n))+γ_DL(list(Γ_n))+γ_REF(list(Γ_n)).

[0071] Semantic Layer (SL) (corresponding to) Figure 2 A unified enterprise data representation methodology for data access. End-user business terminology is allowed using a unified data access point. Effective relationships between different data attributes need to be managed to create a simple business view, providing the most cost-effective business knowledge in the shortest time. It acts as a bridge between raw databases and reporting tools, simplifying self-service reporting, supporting 80% of data and assisting the remaining 20%.

[0072] Data Selection (GCM-IFF): Based on the SL data, the GCM-IFF algorithm is used to select the optimal features. This algorithm combines the exploration and utilization balance of the FFO algorithm, which discovers new solutions during the search process (through the mechanisms of seed dispersal in the field and animal propagation), overcoming the premature convergence problem caused by the random selection in the traditional FFO algorithm. Furthermore, it leverages the GCM algorithm's ability to utilize past successful solutions to generate new solutions. See step 3 and the algorithm pseudocode in the invention description section for details.

[0073] Dynamic App Ontology Construction (corresponding to) Figure 3 ): Select features (speed, memory, bandwidth, cost, turnaround time, waiting time, disk space, burst time, weight, etc.) and ontology components (classes, relations, attributes, axioms, instances), and construct the ontology using Protege. The main processes include: determining feature relations, organizing the hierarchical structure (is-a), defining attributes (object attributes: memory, OS type, network storage, file system, load balancing; data attributes: file size, number of user requests, ID), adding axioms to define the logical relationships between concepts, representing instances, and using HermiT for consistency checks.

[0074] OWL file creation (OwlFilesCreation) (see reference) Figure 4 Use the Protege tool (a free, open-source ontology editor) to convert the created ontology into OWL file format. Protege can be used to create ontology and has built-in reasoning capabilities (such as HermiT), capable of calculating the inferred class hierarchy and checking the consistency of the ontology. The created OWL file contains a hierarchy that machines can use for useful reasoning and to generate RDF instances.

[0075] Cloud Server Interoperability:

[0076] Ontology Manipulation: This involves manipulating the ontology using the HermiTOWL inference engine. HermiTOWL is a hypertableau-based inference engine capable of determining ontology consistency, class containment relationships, and more. The steps are as follows: Set assertions. and DL clause Normalization (reducing complex axioms to axioms containing only classes and attributes, handling SWRL rules, transitive object attributes, and attribute chain axioms), and finally transforming the normalized axioms into DL clauses, using hyperperformance algorithms on the DL clause set. and Abox Perform a consistency check to see Are they consistent?

[0077] Cloud Interoperability Training corresponds to step 7 of the invention content, and includes the algorithm pseudocode of the invention content. Training is performed after the above ontology operations on class features. The Shellgames optimizer is used to optimize the RNN weights and solve the gradient problem. The feature vectors are... As the input layer, the process of calculating the hidden layer and output layer is as follows: Loss function The error is calculated using backpropagation to adjust the weights, and then the weight optimizer using the exchange optimization algorithm is used to update the weights, thus reducing the error. The "shells" represent candidate weight solutions; the exchange optimization algorithm selects the weights, which are the "shells." The "accuracy" and "intelligence" of the weights are then considered, and their positions are adjusted according to the rules for guessing the ball, as can be seen in step 7 of the invention. After training, the model can be used to predict and select the most suitable cloud server.

[0078] Example 2

[0079] Experimental results:

[0080] Experimental results confirm that the present invention has significant advantages:

[0081] High predictive performance: It outperforms DNN, CNN, RNN and I-ANFIS in accuracy (96.32%), precision (94.65%), recall (95.56%), F-score (94.78%), sensitivity (95.32%) and specificity (96.22%).

[0082] High efficiency and low power consumption: The training time was only 22567ms, and the CPU and memory consumption was only 17515kb, which is significantly less than the comparison method.

[0083] Strong robustness: k-fold cross-validation achieved an accuracy of 96.79%-97.99%, far exceeding the average of other models (approximately 92.5%), demonstrating the model's strong generalization ability.

[0084] The superior feature selection GCM-IFF achieves a larger fitness value (102 after 25 iterations) than OSA, BWO, TSA, and GWFA under the same number of iterations, and has a faster and better convergence speed.

[0085] Ontology build efficiency: As the size of the ontology increases, the ontology build time and CPU memory consumption increase linearly. The Protege tool is used for resource management.

[0086] Comparative advantages: Compared with existing representative works [Bassiliades27][Rodriguez-Garcia24][Cimmino30][Adel32][Adel33], the present invention performs better in terms of accuracy (94.65%) and response time (0.019 seconds).

[0087] Those skilled in the art will understand that the embodiments described herein can be modified in various ways without departing from the spirit and scope of this invention. For example, other distributed computing frameworks can be used for data warehouses; other compatible tools besides Protege can be used for ontology construction; feature selection algorithms or neural network structures can also be adjusted according to different scenarios, and the system can also run on other Linux systems. Therefore, the descriptions and figures herein are merely illustrative and should not be considered as limiting the scope of the claims of this invention.

Claims

1. An improved method for semantic interoperability in cloud environments using fusion algorithm recurrent neural networks, comprising the following steps: Data warehouse standardization: Extracting data from applications and cloud servers from the data warehouse, and sequentially performing standardized operations such as data extraction, data cleaning, data transformation, data loading, and data refresh; Semantic layer construction: Import standardized data into the semantic layer, build data attribute relationships, and generate business-oriented data views. Data selection based on GCM-IFF optimization: The optimal features are selected from the semantic layer data using the genetically modified fertile field optimization algorithm (GCM-IFF), with the objective function being: Θ_objective = I_1 Γ_size^D + I_2Γ_TC^D + I_3 Γ_wt^D + I_4 Γ_tat^D + I_5 Γ_mem^c + I_6 Γ_DS^c + I_7 Γ_NR^c; Dynamic application ontology construction: Constructing a dynamic application ontology based on optimal features and implementing consistency checks. OWL file generation: Using the Protege tool to convert the ontology into an OWL file. Cloud server interoperability and ontology operations based on HermiT: Operations on the ontology using the HermiTOWL inferencer, including reading OWL files, checking file consistency, converting OWL into descriptive logical clauses, and performing inference operations based on hyper-performance computing. Cloud interoperability training and prediction: The processed class features are fed into an improved exchange optimization algorithm recurrent neural network for training and testing. The exchange optimization algorithm selects candidate solutions for weights, judges the accuracy and intelligence of the weights, and then updates the position of the weights according to the guessing ball rule, thereby obtaining the interoperability prediction of cloud servers and the optimal cloud server selection.

2. The method according to claim 1, characterized in that, This data warehouse is based on the Apache Hadoop framework and uses the MapReduce processing model to achieve distributed data processing.

3. The method according to claim 1, characterized in that... The semantic layer serves as a bridge connecting the data warehouse and reporting tools, simplifying self-service reporting.

4. The method described in claim 1, characterized in that the GCM-IFF algorithm introduces genetic crossover and mutation during the population update stage, and uses different formulas according to the positive and negative values ​​of the wind dispersion fraction during genetic crossover.

5. The method according to claim 1, characterized in that... Protege is used to build the ontology, and the HermiT inference engine built into Protege is used for consistency checks and inference.

6. The method according to claim 1, characterized in that, The input layer receives class features processed by HermiT, and the hidden and output layers are calculated as follows: X_HL(1) = χ_act( φ_B + ∑∑ w_ij(t)φ^+(t) ) X_HL(2) = χ_act( φ_B + ∑∑ w_ij(t)φ^+(t) ) + X_HL(1) w_ij(t) X_OL(L)(φ_{x,y}) = χ_act^{softmax}( φ_B + ∑∑ w_ij(t)φ^+(t) + X_HL(t-1) ).

7. A cloud environment semantic interoperability system based on an improved exchange optimization algorithm recurrent neural network, comprising: Application Services Module: Provides hosting platform functionality SQL Database Module: Stores application and cloud server information Data warehouse module: Integrates, stores, and processes big data from multiple sources and performs standardized operations. Semantic layer module: Provides unified data interpretation and business view The data selection module is used to run the GCM-IFF algorithm to select the optimal features. Ontology building module: Uses the Protege tool to create dynamic application ontology and generate OWL files. The ontology operation module integrates the HermiT OWL inference engine to enable ontology operation and inference. Cloud interoperability training and prediction module: running model training and optimal cloud server prediction.

8. The system according to claim 7, characterized in that... The data warehouse module uses Apache Hadoop and MapReduce.

9. The system according to claim 7, characterized in that the ontology operation module uses HermiT's "supertable" calculus to perform ontology consistency calculation.

10. The system according to claim 7, characterized in that, The cloud interoperability training and prediction module solves the vanishing and exploding gradient problem by integrating weights using a swap optimization algorithm.

Images