Cloud resource intelligent performance evaluation system and method based on large model

Abstract

The invention relates to a cloud resource intelligent performance evaluation system and method based on a large model, and belongs to the technical field of IT resource management. The system comprises a front-end interaction layer, a knowledge management layer, an AI evaluation layer and a data support layer. The method comprises the following steps: in an application stage, guiding a user to clarify a demand and extract key parameters through multiple rounds of dialogues, automatically matching a calculation formula template and generating a recommendation configuration report; in the review stage, the application document is automatically analyzed, normalization, accuracy and logic compliance verification is carried out based on the knowledge base, and review suggestions are generated. According to the invention, through fusion of the large model, the RAG and other technologies, intelligence and standardization of the resource application and review process are realized, the process efficiency and decision consistency are significantly improved, the operation cost and dependence on expert experience are reduced, and structured precipitation and reuse of enterprise knowledge are realized.

Description

Technical Field

[0001] This invention belongs to the field of IT resource management technology, and in particular to a cloud resource intelligent performance evaluation system and method based on a large model. Background Technology

[0002] The traditional, manually-driven IT resource assessment model involves resource applicants manually drafting application documents, clearly defining the resource type, specifications, and calculation basis before submitting them to the review department. Reviewers then organize cross-departmental offline meetings, relying on expert experience to verify application parameters and calculation logic, referencing internal enterprise resource allocation guidelines, and ultimately providing approval opinions. The core technical measures of this model are "manual document drafting + offline meeting review + expert experience-based decision-making," with information exchanged between stages via offline communication or a simple online work order system.

[0003] Existing technology includes a basic intelligent question-answering system that provides users with basic services such as resource application process queries by searching the knowledge base through keyword matching, but it does not involve resource parameter extraction, formula calculation, or automated review capabilities.

[0004] The existing technology has the following problems: Inefficient: Offline reviews require coordination with experts from multiple departments, resulting in high meeting organization costs and long review cycles for individual tasks, making it unsuitable for high-frequency resource application needs.

[0005] Reliance on expert experience: There is a lack of unified standards for judging the rationality of resource assessments, and the differences in experience among different reviewers lead to inconsistent approval results, and it is difficult to trace historical references.

[0006] High operational threshold: Applicants need to be proficient in the calculation logic and parameter standards of various resources. Without intelligent guidance, problems such as non-standard application documents and missing parameters are likely to occur.

[0007] Manual verification has a large margin of error: reviewers need to manually check the calculation formulas and parameters, which is a lot of work and is prone to misjudgment due to human negligence, resulting in over-application or under-allocation of resources.

[0008] Insufficient knowledge accumulation: Historical application cases and approval experience are stored in the form of unstructured documents, which cannot be reused efficiently, and new reviewers have a long learning curve. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a cloud resource intelligent performance evaluation system and method based on a large model. This addresses the problems of cumbersome processes and low efficiency in traditional methods, enabling online and intelligent resource application and review, and shortening the overall process cycle. It lowers the barrier to resource application by helping applicants quickly generate standardized and reasonable application plans through intelligent guidance and automated calculation. A standardized evaluation system is established to reduce reliance on expert experience and improve the consistency and accuracy of approval results. It enables the structured accumulation and reuse of expert knowledge, forming an iterative resource evaluation knowledge base and parameter template library. It automatically identifies over-configuration or under-configuration issues in resource applications, optimizes resource utilization, and reduces enterprise IT operating costs.

[0010] The technical problem solved by this invention is achieved through the following technical solution: A cloud resource intelligent performance evaluation system based on a large model includes a front-end interaction layer, a knowledge management layer, an AI evaluation layer, and a data support layer. The AI ​​evaluation layer is bidirectionally connected to the front-end interaction layer and the knowledge management layer. The data support layer and the AI ​​evaluation layer bidirectionally interact to evaluate data. The data support layer and the knowledge management layer bidirectionally transfer knowledge construction data and structured knowledge. The system unidirectionally supplies visualized data to the front-end interaction layer and receives operational feedback data. The front-end interaction layer is used to receive resource-related requests from users in natural language and to provide an interactive interface; The knowledge management layer is used to structure historical resource application and review cases and resource allocation specifications to form and maintain a unified knowledge base that includes a parameter template library and a case knowledge base. The AI ​​evaluation layer is used for resource requirement analysis, configuration recommendation, and automated review; the data support layer is used to store a unified knowledge base and support vectorized retrieval of the knowledge base.

[0011] Furthermore, the knowledge management layer includes a data acquisition module, a data cleaning module, a data clustering module, a knowledge storage module, and a fine-tuning dataset generation module. The data acquisition module is used to collect raw data, the data cleaning module is used to standardize the collected raw knowledge data, the data clustering module is used to cluster data according to resource type and business scenario to extract parameter calculation formula templates associated with specific scenarios, the knowledge storage module is used to perform retrieval based on the constructed knowledge base, and the fine-tuning dataset generation module is used to form a large model fine-tuning dataset.

[0012] Furthermore, the AI ​​evaluation layer includes a demand analysis and auxiliary recommendation module, an automated review engine, and a review opinion generation module. The demand analysis and auxiliary recommendation module extracts core business parameters from user input through multi-turn dialogue, retrieves corresponding calculation formula templates from the parameter template library according to resource type, and generates recommended resource configuration schemes through the calculation engine. The automated review engine parses the resource application documents to be reviewed and automatically verifies the standardization, parameter accuracy, and compliance of the calculation logic of the application documents based on a unified knowledge base. When extracting core business parameters, the demand analysis and auxiliary recommendation module adopts a search-enhanced generation method, which assists in parameter extraction or completion by searching similar cases in the case knowledge base. The review opinion generation module generates structured natural language review opinions based on the verification results of the automated review engine.

[0013] Furthermore, the AI ​​evaluation layer also includes a model optimization module, which is used to fine-tune the pre-trained large language model using historical case data and to continuously optimize the model based on user feedback.

[0014] An evaluation method for a cloud resource intelligent performance evaluation system based on a large model includes a resource application method and a resource review method. The resource application method includes the following steps: Step 1: The front-end interaction layer receives the user's initial resource request through natural language input; Step 2: The AI ​​evaluation layer's demand analysis and assisted recommendation module guides users through multiple rounds of dialogue and combines retrieval enhancement generation methods to search the case knowledge base, extracting core business parameters related to resource allocation from user interaction content. Step 3: The demand analysis and auxiliary recommendation module retrieves matching calculation formula templates from the parameter template library of the knowledge management layer based on the identified resource types. Step 4: The demand analysis and auxiliary recommendation module substitutes the extracted core business parameters into the calculation formula template, executes the calculation through the calculation engine, and obtains the recommended resource configuration results. The data support layer synchronously stores the calculation process data. Step 5: The demand analysis and auxiliary recommendation module generates a structured resource request report based on the resource allocation results, and the front-end interaction layer displays the report.

[0015] Furthermore, the resource review method includes the following steps: Step 6: The front-end interaction layer obtains the resource application document uploaded by the user and transmits it to the AI ​​evaluation layer; Step 7: The AI ​​evaluation layer's automated review engine parses the application documents and transforms their content into structured review data; Step 8: The automated review engine calls the approval rule base and parameter template base of the knowledge management layer, and combines the vectorized retrieval capabilities of the data support layer to automatically verify the structured review data. The verification includes at least: verifying the standardization of the application, checking the rationality of the parameters, and reviewing the calculation logic and results. Step 9: The AI ​​evaluation layer's review opinion generation module generates structured natural language review opinions based on the results of automated verification, which are then fed back to the user or reviewer through the front-end interaction layer.

[0016] Furthermore, the enhanced generation in step 2 includes the following steps: Step 2.1, Query Vectorization: Use a text embedding model to convert user input or document content into vectors; Step 2.2, Document Retrieval: Retrieve matching content from the case knowledge base using vector similarity search; Step 2.3, Large Model Enhancement Generation: The input information, search results, parameters and calculation methods are integrated into prompt words, and the large model generates the output results.

[0017] The advantages and positive effects of this invention are: 1. Significantly improved process efficiency: The application cycle for this invention resource has been shortened from an average of 3 working days to within 4 hours, and the review efficiency has been improved by more than 70%. There is no need to organize offline meetings, reducing cross-departmental coordination costs.

[0018] 2. Improved application standardization: Through intelligent guidance and standardized templates, this invention increases the application document completeness compliance rate from 65% to 98%, reducing the rejection rate caused by non-standard documents.

[0019] 3. Rationalization of resource allocation: The automated review engine of this invention can identify more than 85% of over-application scenarios, with an average resource reduction rate of 15%-20%, significantly reducing the waste of enterprise IT resources.

[0020] 4. Enhanced review consistency: Based on a unified parameter template and rule base, this invention increases the consistency of approval results for the same work order from 70% to 95% among different reviewers, avoiding decision-making biases caused by differences in experience.

[0021] 5. Improved knowledge reuse rate: Through the structured accumulation of historical cases and expert experience, this invention shortens the onboarding period for new reviewers from 3 months to 1 week, greatly improving the efficiency of knowledge transfer. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the evaluation system architecture for this invention; Figure 2 This is a schematic diagram of the knowledge management layer architecture of the present invention; Figure 3This is a schematic diagram of the AI ​​evaluation layer and architecture of the present invention; Figure 4 This is a schematic diagram of the interactive window of the present invention; Figure 5 This is a flowchart of the resource application method of the present invention. Detailed Implementation

[0023] The present invention will be further described in detail below with reference to the accompanying drawings.

[0024] A cloud resource intelligent performance evaluation system based on a large model, such as Figure 1 As shown, it includes a front-end interaction layer, a knowledge management layer, an AI evaluation layer, and a data support layer. The AI ​​evaluation layer is connected to the front-end interaction layer and the knowledge management layer. The AI ​​evaluation layer is bidirectionally connected to the front-end interaction layer and the knowledge management layer. The data support layer and the AI ​​evaluation layer bidirectionally interact to evaluate data. The data support layer and the knowledge management layer bidirectionally transfer knowledge construction data and structured knowledge. The data support layer and the front-end interaction layer unidirectionally supply visualized data and receive operational feedback data.

[0025] The front-end interaction layer provides a unified intelligent question-and-answer window, supporting targeted selection of agents (such as resource requirement assessment agents and AI meeting minutes agents), natural language input and multi-turn dialogue completion, and structured output of analysis results.

[0026] The knowledge management layer is used to collect, clean, cluster, store, and update expert knowledge, providing data support for the evaluation process.

[0027] The AI ​​evaluation layer is used for resource requirement analysis, configuration recommendations, and automated review.

[0028] The data support layer consists of a case knowledge base, a parameter template library, and an approval rule base, and achieves efficient retrieval through vector indexing.

[0029] like Figure 2 As shown, the knowledge management layer includes a data acquisition module, a data cleaning module, a data clustering module, a knowledge storage module, and a fine-tuning dataset generation module.

[0030] The data acquisition module is used to parse historical resource application cases, approval opinions, resource allocation specification documents, etc., extract structured information through LLM prompts, and output the original case library and application specifications in JSON format.

[0031] The data cleaning module uses a hybrid extraction method of LLM and regular expressions to standardize resource requirement cases, derivation parameters, and resource type allocation specifications. After manual annotation and proofreading, it outputs high-quality data.

[0032] The data clustering module is used to perform preliminary classification according to resource type (application server, database, etc.), and to achieve scenario segmentation by combining text vectorization and semantic clustering, extracting common parameter formula templates and associating them with applicable scenario tags.

[0033] The knowledge storage module is used to build a unified knowledge base, which includes a case knowledge base, a parameter template library, and an approval rule library. It uses Embeddings for vectorization and builds a vector index to support efficient retrieval.

[0034] The fine-tuning dataset generation module is used to convert historical cases into question-and-answer pairs such as demand analysis, calculation, and approval rationality judgment. Through data augmentation techniques such as synonym rewriting and multi-dimensional question-and-answer generation, a large model fine-tuning dataset is formed.

[0035] like Figure 3 As shown, the AI ​​evaluation layer includes a requirements analysis and auxiliary recommendation module, an automated review engine, and a review opinion generation module. The requirements analysis and auxiliary recommendation module extracts core business parameters from user input through multi-turn dialogue, retrieves corresponding calculation formula templates from the parameter template library based on resource type, and generates recommended resource configuration schemes through the calculation engine. The automated review engine parses application documents and converts them into structured JSON, automatically verifying the requirements in terms of standardization, accuracy, and compliance, reviewing resource calculation logic, and outputting discrepancy prompts. When extracting core business parameters, the requirements analysis and auxiliary recommendation module uses a search-enhanced generation method, retrieving similar cases from the case knowledge base to assist in parameter extraction or completion. The review opinion generation module converts the verification results into natural language suggestions for reviewers to confirm or adjust.

[0036] Process Q&A guidance, historical case Q&A, and resource application assistance recommendations are all core functions of the AI ​​evaluation layer's requirements analysis and assistance recommendation module: Process Q&A guidance relies on the process knowledge base of the knowledge management layer and generates operation guidelines through RAG technology; Historical case Q&A recalls similar cases through the Milvus vector database of the data support layer; Resource application assistance recommendations collect parameters through multi-round dialogues, call parameter template library formulas, and combine the Sympy calculation engine to complete configuration calculations, with data storage and retrieval support provided by the data support layer throughout the process.

[0037] The AI ​​evaluation layer also includes a model optimization module, which is used to fine-tune the pre-trained large language model using historical case data and continuously optimize the model based on user feedback.

[0038] Baseline model selection: Qwen32B is adopted as the baseline model, and the lightweight fine-tuning or knowledge distillation optimization path is selected according to computing resources.

[0039] Training strategy: Ray is used as the distributed engine, combined with ColossalAI or deepspeed to support parallel training. Based on open source projects such as LLaMAFactory, LoRA is fine-tuned by setting hyperparameters such as num_epochs=3 and lr=2e-5.

[0040] Knowledge Updates: The knowledge base is automatically updated by adding new resource application cases. The satisfaction rate of Q&A is analyzed based on user feedback (likes / dislikes). When the deviation rate increases, the model is incrementally fine-tuned.

[0041] An evaluation method for a cloud resource intelligent performance evaluation system based on a large model includes resource application methods and resource review methods, such as... Figure 5 As shown, the resource application method includes the following steps: Step 1: The front-end interaction layer receives the user's initial resource request through natural language input.

[0042] Step 2: The AI ​​evaluation layer's demand analysis and assisted recommendation module guides users through multiple rounds of dialogue and uses a search enhancement generation method to retrieve case knowledge bases, extracting core business parameters related to resource allocation from user interaction content.

[0043] Search enhancement generation includes the following steps: Step 2.1, Query Vectorization: Use a text embedding model to convert user input or document content into vectors.

[0044] Step 2.2, Document Retrieval: Retrieve matching content from the case knowledge base using vector similarity search.

[0045] Step 2.3, Large Model Enhancement Generation: The input information, search results, parameters and calculation methods are integrated into prompt words, and the large model generates the output results.

[0046] Retrieval Enhancement Generation (RAG) Architecture: It adopts a three-level structure of text embedding model + Milvus vector database + large model. The text embedding model converts unstructured data into high-dimensional vectors, Milvus database provides millisecond-level vector similarity retrieval, and large model integrates input and retrieval results to generate output. Large model optimization architecture: Based on Qwen32B as the baseline model, LoRA lightweight fine-tuning technology is used to enhance the ability of specific tasks. The training architecture adopts the ray distributed engine and combines ColossalAI or deepspeed to achieve data parallelism, model parallelism and pipeline parallelism. The underlying architecture is based on the PyTorch framework and relies on open source projects such as LLaMAFactory for modification and adaptation. Calculation engine algorithm: Integrates Sympy symbolic computation library, supports linear calculations, proportional conversions and other mathematical operations related to resource configuration, and ensures accurate calculation after parameter substitution.

[0047] The core processing flow is as follows: Query vectorization: Using a text embedding model to convert user-input questions into vectors; Document retrieval: Using vector similarity search to retrieve similar cases in the case knowledge base or to detect the parameter information needed to estimate the amount of resources in a resource application; Large model augmentation generation: The user input, retrieved cases, derivation parameters, and calculation methods are combined to form RAG prompt words, which are then output by the large model.

[0048] Model optimization Prioritize using cue word engineering to evaluate model capabilities. When the model cannot effectively complete the task, fine-tuning is needed to enhance the model's vertical capabilities.

[0049] 1. Fine-tune the goals: Capability 1: Parameter Extraction and Calculation. It can extract core parameters such as "concurrency and log growth rate" from natural language and apply formulas to calculate them.

[0050] Capability 2: Approval evaluation reasoning can combine resource application criteria to determine whether an application exceeds the specifications or wastes resources, and provide reasons and recommended configurations.

[0051] 2. Training process: (1) Prepare the training dataset (thousands of instruction-response pairs); (2) LoRA fine-tuning; (3) Evaluate the model’s performance: Use a set of reserved cases to test the model’s reasoning accuracy and formula consistency.

[0052] 3. Baseline Model: Qwen32B was adopted as the baseline model for evaluation and capability verification. Based on its performance in tasks such as resource request inference, parameter extraction, and review opinion generation, the adaptability of the large model to the current business scenario was evaluated. Based on the evaluation, and considering actual deployment requirements such as inference performance and computing resource availability, the following optimization paths were selected: (1) Sufficient computing resources allow for lightweight fine-tuning of the baseline model; (2) When computing resources are scarce, knowledge distillation can be used to transfer knowledge to a model with fewer parameters.

[0053] 4. Model training strategy: Using ray as the distributed engine, it calls container resources for memory optimization, data parallelism, pipeline parallelism, and model parallelism strategies supported by ColossalAI or deepspeed. Training fine-tuning is based on open-source projects such as LLaMAFactory, ChatLLaMA, and FireFly, which are modified to adapt loss functions, optimizers, and training strategy algorithms. The model library functionality is supported by the hugging-face module, and the underlying deep learning framework is PyTorch.

[0054] Step 3: The requirements analysis and auxiliary recommendation module retrieves matching calculation formula templates from the parameter template library of the knowledge management layer based on the identified resource types.

[0055] Step 4: The demand analysis and auxiliary recommendation module substitutes the extracted core business parameters into the calculation formula template, executes the calculation through the calculation engine, and obtains the recommended resource configuration results. The data support layer synchronously stores the calculation process data.

[0056] Step 5: The demand analysis and auxiliary recommendation module generates a structured resource request report based on the resource allocation results, and the front-end interaction layer displays the report.

[0057] The resource review methodology includes the following steps: Step 6: The front-end interaction layer obtains the resource application document uploaded by the user and transmits it to the AI ​​evaluation layer.

[0058] Step 7: The AI ​​evaluation layer's automated review engine parses the application documents and transforms their content into structured review data.

[0059] Step 8: The automated review engine calls the approval rule library and parameter template library of the knowledge management layer, and combines the vectorized retrieval capabilities of the data support layer to automatically verify the structured review data. The verification includes at least: verifying the standardization of the application, checking the rationality of the parameters, and reviewing the calculation logic and results.

[0060] Step 9: The AI ​​evaluation layer's review opinion generation module generates structured natural language review opinions based on the results of automated verification, which are then fed back to the user or reviewer through the front-end interaction layer.

[0061] In the implementation of resource application and resource review methods, a search-enhanced generation method is adopted to obtain similar cases or standard rules by querying the case knowledge base to assist in the processing.

[0062] Based on the above-mentioned intelligent performance evaluation system and method for cloud resources based on a large model, the effectiveness of the present invention is verified by the results of the following two embodiments.

[0063] Example 1: Application server resource request: Application scenario: A business is requesting application server resources.

[0064] Key parameters: 1000 concurrent users, 30 backend services, 27 frontend services, 1GB / day log growth, 180 days log retention, 20% redundancy policy.

[0065] Processing procedure: The user selects a resource requirement assessment agent and inputs business requirements. The agent confirms the above parameters through multiple rounds of dialogue, calls the "Web application computing template", and calculates according to the formulas "CPU = 0.4 × number of backend services + 0.1 × number of frontend services" and "storage = 20 + log growth × log retention days + 100".

[0066] Output: Recommended configuration: 14 CPU cores (0.4×30+0.1×27=14.7, rounded down to 14), 52.5GB RAM (1.5×30+0.5×27=52.5), 200GB storage (20+1×180+100=300, 250GB recommended after considering 20% ​​redundancy), generating a structured application report.

[0067] Example 2: Figure 4 As shown, ClickHouse data warehouse resource review: Application information: Write 220 million data entries per day (1.7KB each), retain data for 180 days, dual replicas, compression ratio 0.2, expansion factor 1.2, maximum disk utilization 75%, request storage resources of 50TB.

[0068] Review process: After the automated review engine parses the application documents, it calls the storage calculation formula "Storage = (Daily raw data volume × Number of copies × Retention days × Compression ratio × Expansion coefficient) ÷ Disk utilization rate", which calculates to 40TB (360GB × 2 × 180 × 0.2 × 1.2 ÷ 0.75 = 40TB).

[0069] Review result: The requested storage of 50TB exceeds the reasonable configuration. The review opinion is: "The current storage request is 50TB. After review by the formula, the reasonable configuration is 40TB. It is recommended to apply for 40TB, or provide additional explanations of the additional storage needs."

[0070] It should be emphasized that the embodiments described in this invention are illustrative rather than limiting. Therefore, this invention includes, but is not limited to, the embodiments described in the specific implementation. Any other implementations derived by those skilled in the art based on the technical solutions of this invention are also within the scope of protection of this invention.

Claims

1. A cloud resource intelligent performance evaluation system based on a large model, characterized in that, It includes a front-end interaction layer, a knowledge management layer, an AI evaluation layer, and a data support layer. The AI ​​evaluation layer is bidirectionally connected to the front-end interaction layer and the knowledge management layer. The data support layer and the AI ​​evaluation layer interact bidirectionally to evaluate data. The data support layer and the knowledge management layer exchange knowledge construction data and structured knowledge bidirectionally. The data support layer and the front-end interaction layer provide visualized data and receive operational feedback data unidirectionally. The front-end interaction layer is used to receive resource-related requests from users in natural language and to provide an interactive interface; The knowledge management layer is used to structure historical resource application and review cases and resource allocation specifications to form and maintain a unified knowledge base that includes a parameter template library and a case knowledge base. The AI ​​evaluation layer is used for resource requirement analysis, configuration recommendation, and automated review; the data support layer is used to store a unified knowledge base and support vectorized retrieval of the knowledge base.

2. The cloud resource intelligent performance evaluation system based on a large model according to claim 1, characterized in that, The knowledge management layer includes a data acquisition module, a data cleaning module, a data clustering module, a knowledge storage module, and a fine-tuning dataset generation module. The data acquisition module is used to collect raw data, the data cleaning module is used to standardize the collected raw knowledge data, the data clustering module is used to cluster data according to resource type and business scenario to extract parameter calculation formula templates associated with specific scenarios, the knowledge storage module is used to perform retrieval based on the constructed knowledge base, and the fine-tuning dataset generation module is used to form a large model fine-tuning dataset.

3. The cloud resource intelligent performance evaluation system based on a large model according to claim 1, characterized in that, The AI ​​evaluation layer includes a demand analysis and auxiliary recommendation module, an automated review engine, and a review opinion generation module. The demand analysis and auxiliary recommendation module is used to extract core business parameters from user input through multi-round dialogue, retrieve corresponding calculation formula templates from the parameter template library according to resource type, and generate recommended resource configuration schemes through the calculation engine. The automated review engine parses resource application documents to be reviewed and automatically verifies the standardization, parameter accuracy, and compliance of calculation logic of the application documents based on a unified knowledge base. The requirements analysis and auxiliary recommendation module uses a search-enhanced generation method when extracting core business parameters. It assists in parameter extraction or completion by searching for similar cases in the case knowledge base. The review opinion generation module generates structured natural language review opinions based on the verification results of the automated review engine.

4. The cloud resource intelligent performance evaluation system based on a large model according to claim 1, characterized in that, The AI ​​evaluation layer also includes a model optimization module, which is used to fine-tune the pre-trained large language model using historical case data and to continuously optimize the model based on user feedback.

5. An evaluation method for a cloud resource intelligent performance evaluation system based on a large model as described in any one of claims 1 to 4, characterized in that, This includes resource application methods and resource review methods. The resource application method includes the following steps: Step 1: The front-end interaction layer receives the user's initial resource request through natural language input; Step 2: The AI ​​evaluation layer's demand analysis and assisted recommendation module guides users through multiple rounds of dialogue and combines retrieval enhancement generation methods to search the case knowledge base, extracting core business parameters related to resource allocation from user interaction content. Step 3: The demand analysis and auxiliary recommendation module retrieves matching calculation formula templates from the parameter template library of the knowledge management layer based on the identified resource types. Step 4: The demand analysis and auxiliary recommendation module substitutes the extracted core business parameters into the calculation formula template, executes the calculation through the calculation engine, and obtains the recommended resource configuration results. The data support layer synchronously stores the calculation process data. Step 5: The demand analysis and auxiliary recommendation module generates a structured resource request report based on the resource allocation results, and the front-end interaction layer displays the report.

6. The evaluation method for a cloud resource intelligent performance evaluation system based on a large model according to claim 5, characterized in that, The resource review method includes the following steps: Step 6: The front-end interaction layer obtains the resource application document uploaded by the user and transmits it to the AI ​​evaluation layer; Step 7: The AI ​​evaluation layer's automated review engine parses the application documents and transforms their content into structured review data; Step 8: The automated review engine calls the approval rule base and parameter template base of the knowledge management layer, and combines the vectorized retrieval capabilities of the data support layer to automatically verify the structured review data. The verification includes at least: verifying the standardization of the application, checking the rationality of the parameters, and reviewing the calculation logic and results. Step 9: The AI ​​evaluation layer's review opinion generation module generates structured natural language review opinions based on the results of automated verification, which are then fed back to the user or reviewer through the front-end interaction layer.

7. The evaluation method for a cloud resource intelligent performance evaluation system based on a large model according to claim 6, characterized in that, Step 2, the retrieval enhancement generation, includes the following steps: Step 2.1, Query Vectorization: Use a text embedding model to convert user input or document content into vectors; Step 2.2, Document Retrieval: Retrieve matching content from the case knowledge base using vector similarity search; Step 2.3, Large Model Enhancement Generation: The input information, search results, parameters and calculation methods are integrated into prompt words, and the large model generates the output results.

Images