Virtual machine resource opening method and system combined with multi-modal risk assessment

Abstract

The invention provides a virtual machine resource opening method and system combined with multi-modal risk assessment in the technical field of virtual machine management, and the method comprises the steps: S1, creating an order semantic recognition model, a risk assessment model and a resource dynamic distribution model, and carrying out the training and deployment; s2, collecting and calculating resource data and constructing a resource thermodynamic diagram; s3, obtaining a real-time virtual machine purchase order, performing verification and analysis to obtain order content and a user account, and obtaining real-time multi-modal data based on the user account; s4, inputting the order content into the order semantic recognition model to obtain a configuration demand, and inputting the real-time multi-modal data into the risk assessment model to obtain a risk assessment result; and S5, based on the risk assessment result, the configuration demand and the resource thermodynamic diagram, calling a resource dynamic allocation model to execute resource allocation. The method has the advantages that the virtual machine resource opening efficiency, the resource utilization rate and the safety are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual machine management, and in particular to a method and system for provisioning virtual machine resources combined with multimodal risk assessment. Background Art

[0002] As a core component of cloud computing infrastructure, virtual machine technology builds isolated virtual computing environments on physical hosts through the hypervisor virtualization layer. This technology, thanks to its environmental isolation, resource configurability, and deployment flexibility, has fostered a diverse application ecosystem: in development and testing, it supports the parallel development of multiple versions of environments; in education and research, it enables cross-platform operating system experiments; in enterprise IT, it pools and consolidates server resources, increasing hardware utilization by 40%-60%; and for end users, it provides cross-platform application compatibility solutions.

[0003] Commercial virtual machines are typically deployed on cloud platforms (servers). Virtual machines are created by purchasing services from corresponding service providers, and then the virtual machines call upon the cloud platform's computing resources to perform related computing tasks. The virtual machine service purchase process first requires submitting a purchase order to the cloud platform. Upon receiving the order, the cloud platform reviews it and only activates the corresponding virtual machine resources based on the order. Traditionally, manual review and activation procedures are required, but manual review is time-consuming and labor-intensive, and can lead to errors that could affect virtual machine activation. This process no longer meets current needs, leading to the need for automated review and activation of virtual machine resources.

[0004] However, the existing automated review and activation system has the following problems: 1. It adopts predefined threshold strategies (such as automatic approval for credit scores > 80), but lacks natural language processing capabilities and cannot identify semantic conflicts in order texts (for example, applying for a "financial-grade security environment" but checking the "public network policy"). This leads to a high system misjudgment rate, forcing operation and maintenance personnel to frequently intervene in rule adjustments, which in turn affects the efficiency of virtual machine resource activation; 2. It adopts a linear allocation mechanism of FIFO queues, which is prone to resource fragmentation in sudden high-concurrency request scenarios (such as the remaining 36-core CPU is split into three 12-core nodes) and causes regional load imbalance; 3. It activates resources based solely on virtual machine purchase orders, without combining multimodal data (historical orders, behavioral data, etc.) for comprehensive consideration and risk assessment. This leads to problems such as improper resource allocation, and the order may be tampered with during transmission, causing the virtual machine to crash during actual operation due to inability to undertake the corresponding load.

[0005] Therefore, how to provide a virtual machine resource provisioning method and system that combines multimodal risk assessment to improve the efficiency, resource utilization and security of virtual machine resource provisioning has become a technical problem that needs to be solved urgently. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for virtual machine resource provisioning combined with multimodal risk assessment, so as to improve the efficiency, resource utilization and security of virtual machine resource provisioning.

[0007] In a first aspect, the present invention provides a method for provisioning virtual machine resources in combination with multimodal risk assessment, comprising the following steps: Step S1: The cloud platform creates an order semantic recognition model, a risk assessment model, and a resource dynamic allocation model; Step S2: The cloud platform obtains a large amount of historical multimodal data including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs, and constructs a data set after preprocessing and annotating each of the historical multimodal data; Step S3: The cloud platform trains the order semantic recognition model, the risk assessment model, and the resource dynamic allocation model using the data set, and deploys the trained order semantic recognition model, the risk assessment model, and the resource dynamic allocation model; Step S4: The cloud platform collects computing resource data in real time and constructs a resource heat map based on the collected computing resource data; Step S5: The cloud platform obtains a real-time virtual machine purchase order, verifies and parses the real-time virtual machine purchase order to obtain order content and a user account, and obtains real-time multimodal data based on the user account; Step S6: The cloud platform inputs the order content into the deployed order semantic recognition model to obtain configuration requirements, and inputs the real-time multimodal data into the risk assessment model to obtain risk assessment results; Step S7: The cloud platform calls the resource dynamic allocation model to allocate computing resources based on the risk assessment results, configuration requirements, and resource heat map, so as to perform a virtual machine resource provisioning operation; Step S8: The cloud platform feeds back the activation result and records the real-time activation log; In step S1, the order semantic recognition model is constructed based on an input representation module, a context encoding module, a multi-task decoding module, and a post-processing enhancement module; the input representation module is used to convert the virtual machine purchase order into an order vector; the context encoding module is constructed based on a BiLSTM network and a Transformer network, the BiLSTM network is used to extract local dependency features from the order vector, and the Transformer network is used to extract global dependency features from the order vector; the multi-task decoding module is used to extract demand parameters, demand strength, and implicit demand from the local dependency features and the global dependency features; the post-processing enhancement module is used to perform semantic normalization on the demand parameters, demand strength, and implicit demand to obtain configuration requirements; The risk assessment model is constructed based on a multimodal input module, a feature extraction module, a feature fusion module, and a risk output module; the multimodal input module is used to input multimodal data; the feature extraction module is used to extract multimodal features including semantic conflict level, abnormal operation frequency, geographical distribution of request sources, and resource over-allocation ratio from the multimodal data; the feature fusion module is used to fuse the multimodal features through a random forest classifier to obtain a fused feature; the risk output module is used to output a risk assessment result with a risk level based on the fused feature; the risk level is low risk, medium risk, or high risk; The dynamic resource allocation model is built based on a dynamic graph neural network and is used to generate resource allocation plans based on configuration requirements and resource heat maps.

[0008] Furthermore, the step S2 is specifically as follows: The cloud platform acquires a large amount of historical multimodal data including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs; the historical virtual machine purchase orders include at least the order number, order content, purchase time, purchaser information, and service terms; the historical purchase behavior data includes at least user attributes, order time, purchase frequency, payment method, billing method, IP address, purchase channel, browsing behavior, and interaction behavior; the historical resource data includes at least CPU utilization, memory utilization, storage utilization, storage growth trend, network traffic, network type, network bandwidth, and network topology; the historical activation log includes at least transaction records, activation time, virtual machine name, virtual machine configuration, and user account; Performing preprocessing on each of the historical multimodal data, including at least stop word removal, outlier removal, missing value filling, word segmentation, and format unification; The pre-processed historical multimodal data are labeled with configuration requirements, risk levels, resource heat maps, and resource allocation plans, and a data set is constructed based on the labeled historical multimodal data.

[0009] Furthermore, the step S3 is specifically as follows: The cloud platform divides the dataset into a training set, a validation set, and a test set based on a ratio of 8:1:1. The order semantic recognition model, the risk assessment model, and the dynamic resource allocation model are trained using the corresponding data in the dataset. During the training process, the hyperparameters including at least the learning rate, the learning decay rate, the random dropout rate, the batch size, and the number of iterations are continuously optimized. The weights and biases of the order semantic recognition model, the risk assessment model, and the dynamic resource allocation model are updated in combination with the backpropagation algorithm until the preset convergence conditions are met. The trained order semantic recognition model, risk assessment model and resource dynamic allocation model are verified in turn through the verification set, and the verified order semantic recognition model, risk assessment model and resource dynamic allocation model are tested in turn through the test set, and the tested order semantic recognition model, risk assessment model and resource dynamic allocation model are converted into ONNX format and deployed.

[0010] Furthermore, the step S4 is specifically as follows: The cloud platform collects computing resource data in real time, including at least CPU utilization, memory utilization, storage utilization, storage growth trend, network traffic, network type, network bandwidth, and network topology; performs preprocessing on each computing resource data, including at least cleaning, conversion, and aggregation; maps the preprocessed computing resource data based on a preset display method to obtain a resource heat map; and updates the resource heat map through a preset sliding time window; The step S5 is specifically as follows: The cloud platform obtains a real-time virtual machine purchase order carrying an encrypted message, a hardware fingerprint, a timestamp, and a hash value; the encrypted message is obtained by encrypting the order content and the user account, and the specific encryption rules are as follows: encrypting the order content and the user account into a first layer of encrypted data using the AES algorithm, converting the first layer of encrypted data into hexadecimal data, swapping the number 6 and the letter A, and the number 9 and the letter B in the hexadecimal data to obtain a second layer of encrypted data, encrypting the second layer of encrypted data into a third layer of encrypted data using the 3DES algorithm, shifting each character of the third layer of encrypted data right by 8 bits to obtain a fourth layer of encrypted data, and encrypting the fourth layer of encrypted data into an encrypted message using the RSA algorithm; the hardware fingerprint is obtained by concatenating the device serial number and the MAC address; the hash value is obtained by hashing the encrypted message, the hardware fingerprint, and the timestamp; The cloud platform parses the received real-time virtual machine purchase order to obtain an encrypted message, a hardware fingerprint, a timestamp, and a hash value. After performing an integrity check using the hash value, an aging check is performed using the timestamp, and then a legitimacy check is performed using the hardware fingerprint. The encrypted message is then decrypted into four layers of encrypted data using the RSA algorithm, and each character of the four layers of encrypted data is circularly shifted to the left by 8 bits to obtain three layers of encrypted data. The three layers of encrypted data are decrypted into two layers of encrypted data using the 3DES algorithm, and the number 6 and the letter A in the two layers of encrypted data are swapped, and the number 9 and the letter B are swapped to obtain hexadecimal data. The hexadecimal data is inversely converted to obtain a layer of encrypted data. The layer of encrypted data is decrypted into order content and a user account using the AES algorithm, and real-time multimodal data is obtained based on the user account.

[0011] Furthermore, the step S7 is specifically as follows: The cloud platform analyzes the risk assessment result. When the risk assessment result is low risk, the configuration requirements and the resource heat map are input into the resource dynamic allocation model to obtain a resource allocation plan, and the virtual machine resource provisioning operation is automatically performed based on the resource allocation plan. When the risk assessment result is medium risk, the configuration requirements and the resource heat map are input into the resource dynamic allocation model to obtain a resource allocation plan, and the resource allocation plan is encrypted and pushed to a pre-associated management terminal in real time for approval. When the risk assessment result is high risk, the virtual machine resource provisioning operation is refused. The step S8 is specifically as follows: The cloud platform generates an activation result indicating activation success or failure, encrypts the activation result into first-level encrypted data using the RC6 algorithm, splits the first-level encrypted data into second-level encrypted data at a ratio of 1:4 and swaps the order of the first and second-level encrypted data, encrypts the second-level encrypted data into an encrypted notification using the Twofish algorithm, and feeds back the encrypted notification. The cloud platform records real-time activation logs, encrypts the real-time activation logs into first encrypted data using the IDEA algorithm, adds a random string of a specified length to a specified position of the first encrypted data to obtain second encrypted data, encrypts the second encrypted data into an encrypted log using the SM4 algorithm, and stores and distributes the encrypted logs.

[0012] In a second aspect, the present invention provides a virtual machine resource provisioning system combined with multimodal risk assessment, comprising the following modules: The model creation module is used for the cloud platform to create an order semantic recognition model, a risk assessment model, and a resource dynamic allocation model; A dataset construction module is used for the cloud platform to obtain a large amount of historical multimodal data, including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs, and to construct a dataset after preprocessing and annotating each of the historical multimodal data; A model training module is used for the cloud platform to train the order semantic recognition model, risk assessment model and resource dynamic allocation model through the data set, and deploy the trained order semantic recognition model, risk assessment model and resource dynamic allocation model; A resource heat map construction module is used for collecting computing resource data in real time on the cloud platform and constructing a resource heat map based on the collected computing resource data; A purchase order acquisition module is used for the cloud platform to obtain real-time virtual machine purchase orders, verify and parse the real-time virtual machine purchase orders to obtain order content and user accounts, and obtain real-time multimodal data based on the user accounts; A semantic recognition risk assessment module is used for the cloud platform to input the order content into the deployed order semantic recognition model to obtain configuration requirements, and input the real-time multimodal data into the risk assessment model to obtain risk assessment results; A virtual machine resource provisioning module is used for the cloud platform to call the resource dynamic allocation model to allocate computing resources based on the risk assessment results, configuration requirements and resource heat map, so as to perform virtual machine resource provisioning operations; The activation result feedback module is used to feedback the activation results to the cloud platform and record the real-time activation log; In the model creation module, the order semantic recognition model is constructed based on an input representation module, a context encoding module, a multi-task decoding module, and a post-processing enhancement module; the input representation module is used to convert a virtual machine purchase order into an order vector; the context encoding module is constructed based on a BiLSTM network and a Transformer network, the BiLSTM network is used to extract local dependency features from the order vector, and the Transformer network is used to extract global dependency features from the order vector; the multi-task decoding module is used to extract demand parameters, demand strength, and implicit demand from the local dependency features and the global dependency features; the post-processing enhancement module is used to perform semantic normalization on the demand parameters, demand strength, and implicit demand to obtain configuration requirements; The risk assessment model is constructed based on a multimodal input module, a feature extraction module, a feature fusion module, and a risk output module; the multimodal input module is used to input multimodal data; the feature extraction module is used to extract multimodal features including semantic conflict level, abnormal operation frequency, geographical distribution of request sources, and resource over-allocation ratio from the multimodal data; the feature fusion module is used to fuse the multimodal features through a random forest classifier to obtain a fused feature; the risk output module is used to output a risk assessment result with a risk level based on the fused feature; the risk level is low risk, medium risk, or high risk; The dynamic resource allocation model is built based on a dynamic graph neural network and is used to generate resource allocation plans based on configuration requirements and resource heat maps.

[0013] Furthermore, the dataset construction module is specifically used to: The cloud platform acquires a large amount of historical multimodal data including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs; the historical virtual machine purchase orders include at least the order number, order content, purchase time, purchaser information, and service terms; the historical purchase behavior data includes at least user attributes, order time, purchase frequency, payment method, billing method, IP address, purchase channel, browsing behavior, and interaction behavior; the historical resource data includes at least CPU utilization, memory utilization, storage utilization, storage growth trend, network traffic, network type, network bandwidth, and network topology; the historical activation log includes at least transaction records, activation time, virtual machine name, virtual machine configuration, and user account; Performing preprocessing on each of the historical multimodal data, including at least stop word removal, outlier removal, missing value filling, word segmentation, and format unification; The pre-processed historical multimodal data are labeled with configuration requirements, risk levels, resource heat maps, and resource allocation plans, and a data set is constructed based on the labeled historical multimodal data.

[0014] Furthermore, the model training module is specifically used to: The cloud platform divides the dataset into a training set, a validation set, and a test set based on a ratio of 8:1:1. The order semantic recognition model, the risk assessment model, and the dynamic resource allocation model are trained using the corresponding data in the dataset. During the training process, the hyperparameters including at least the learning rate, the learning decay rate, the random dropout rate, the batch size, and the number of iterations are continuously optimized. The weights and biases of the order semantic recognition model, the risk assessment model, and the dynamic resource allocation model are updated in combination with the backpropagation algorithm until the preset convergence conditions are met. The trained order semantic recognition model, risk assessment model and resource dynamic allocation model are verified in turn through the verification set, and the verified order semantic recognition model, risk assessment model and resource dynamic allocation model are tested in turn through the test set, and the tested order semantic recognition model, risk assessment model and resource dynamic allocation model are converted into ONNX format and deployed.

[0015] Furthermore, the resource heat map construction module is specifically used to: The cloud platform collects computing resource data in real time, including at least CPU utilization, memory utilization, storage utilization, storage growth trend, network traffic, network type, network bandwidth, and network topology; performs preprocessing on each computing resource data, including at least cleaning, conversion, and aggregation; maps the preprocessed computing resource data based on a preset display method to obtain a resource heat map; and updates the resource heat map through a preset sliding time window; The purchase order acquisition module is specifically used to: The cloud platform obtains a real-time virtual machine purchase order carrying an encrypted message, a hardware fingerprint, a timestamp, and a hash value; the encrypted message is obtained by encrypting the order content and the user account, and the specific encryption rules are as follows: encrypting the order content and the user account into a first layer of encrypted data using the AES algorithm, converting the first layer of encrypted data into hexadecimal data, swapping the number 6 and the letter A, and the number 9 and the letter B in the hexadecimal data to obtain a second layer of encrypted data, encrypting the second layer of encrypted data into a third layer of encrypted data using the 3DES algorithm, shifting each character of the third layer of encrypted data right by 8 bits to obtain a fourth layer of encrypted data, and encrypting the fourth layer of encrypted data into an encrypted message using the RSA algorithm; the hardware fingerprint is obtained by concatenating the device serial number and the MAC address; the hash value is obtained by hashing the encrypted message, the hardware fingerprint, and the timestamp; The cloud platform parses the received real-time virtual machine purchase order to obtain an encrypted message, a hardware fingerprint, a timestamp, and a hash value. After performing an integrity check using the hash value, an aging check is performed using the timestamp, and then a legitimacy check is performed using the hardware fingerprint. The encrypted message is then decrypted into four layers of encrypted data using the RSA algorithm, and each character of the four layers of encrypted data is circularly shifted to the left by 8 bits to obtain three layers of encrypted data. The three layers of encrypted data are decrypted into two layers of encrypted data using the 3DES algorithm, and the number 6 and the letter A in the two layers of encrypted data are swapped, and the number 9 and the letter B are swapped to obtain hexadecimal data. The hexadecimal data is inversely converted to obtain a layer of encrypted data. The layer of encrypted data is decrypted into order content and a user account using the AES algorithm, and real-time multimodal data is obtained based on the user account.

[0016] Furthermore, the virtual machine resource provisioning module is specifically used to: The cloud platform analyzes the risk assessment result. When the risk assessment result is low risk, the configuration requirements and the resource heat map are input into the resource dynamic allocation model to obtain a resource allocation plan, and the virtual machine resource provisioning operation is automatically performed based on the resource allocation plan. When the risk assessment result is medium risk, the configuration requirements and the resource heat map are input into the resource dynamic allocation model to obtain a resource allocation plan, and the resource allocation plan is encrypted and pushed to a pre-associated management terminal in real time for approval. When the risk assessment result is high risk, the virtual machine resource provisioning operation is refused. The activation result feedback module is specifically used to: The cloud platform generates an activation result indicating activation success or failure, encrypts the activation result into first-level encrypted data using the RC6 algorithm, splits the first-level encrypted data into second-level encrypted data at a ratio of 1:4 and swaps the order of the first and second-level encrypted data, encrypts the second-level encrypted data into an encrypted notification using the Twofish algorithm, and feeds back the encrypted notification. The cloud platform records real-time activation logs, encrypts the real-time activation logs into first encrypted data using the IDEA algorithm, adds a random string of a specified length to a specified position of the first encrypted data to obtain second encrypted data, encrypts the second encrypted data into an encrypted log using the SM4 algorithm, and stores and distributes the encrypted logs.

[0017] The advantages of the present invention are: 1. Create an order semantic recognition model, a risk assessment model, and a resource dynamic allocation model through the cloud platform, obtain a large amount of historical multimodal data including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs, pre-process and annotate each historical multimodal data to build a data set, and use the data set to train and deploy the order semantic recognition model, risk assessment model, and resource dynamic allocation model; then the cloud platform collects computing resource data in real time to build a resource heat map, verifies and parses the obtained real-time virtual machine purchase orders to obtain order content and user accounts, obtains real-time multimodal data based on the user account, inputs the order content into the deployed order semantic recognition model to obtain configuration requirements, and inputs the real-time multimodal data into the risk assessment model to obtain risk assessment results; then, based on the risk assessment results, configuration requirements, and resource heat map, calls resources. The dynamic allocation model executes the allocation of computing resources to execute the virtual machine resource activation operation. The cloud platform feedbacks the activation results and records the real-time activation log. That is, the virtual machine resource activation operation is executed based on the pre-trained order semantic recognition model, risk assessment model and resource dynamic allocation model. When the risk assessment result is low risk, the entire process is automatic and no human intervention is required. The order semantic recognition model can effectively identify semantic conflicts in order texts, improve the accuracy of configuration requirement identification, and avoid human intervention as much as possible. The resource heat map can effectively integrate resource fragments, and combine with the resource dynamic allocation model to dynamically allocate computing resources, which can effectively utilize computing resources. In addition, during the virtual machine resource activation process, the real-time virtual machine purchase order is verified to prevent the order from being tampered with. Combined with the risk assessment of the risk assessment model, the efficiency, resource utilization and security of virtual machine resource activation are ultimately greatly improved.

[0018] 2. By setting up an order semantic recognition model based on the input representation module, context encoding module, multi-task decoding module and post-processing enhancement module, combined with hierarchical feature extraction (local to global), multi-task joint optimization (explicit demand parameters and demand intensity + implicit demand), heterogeneous model fusion (BiLSTM + Transformer) and output normalization, deep semantic analysis of virtual machine purchase orders is achieved, thereby greatly improving the accuracy of configuration demand identification.

[0019] 3. The feature extraction module of the risk assessment model focuses on four major features: semantic conflict level, abnormal operation frequency, geographical distribution, and resource over-allocation ratio. It covers key dimensions such as risk behavior pattern (abnormal frequency), intent recognition (semantic conflict), environmental anomalies (geographic anomalies), and system vulnerability (resource over-allocation). It has clear business explanations. For example, geographical distribution anomalies can identify database stuffing attacks, and resource over-allocation can warn of system overload risks. It can comprehensively assess the risks in the process of virtual machine resource activation, thereby greatly improving the security of virtual machine resource activation.

[0020] 4. By using a dynamic graph neural network to analyze resource heat maps in real time (based on real-time indicators such as CPU, memory, and network), combined with a sliding time window update strategy, flexible resource allocation scheduling is achieved. Compared with static allocation strategies, the utilization rate of computing resources is greatly improved.

[0021] 5. By designing a multi-layer hybrid encryption mechanism (AES+3DES+RSA+character replacement) in real-time virtual machine purchase orders, combined with hardware fingerprint (device serial number + MAC address), timestamp, and hash value verification, it can resist man-in-the-middle attacks and replay attacks, and reduce the cost of cracking encrypted messages to the order of 2^128, bringing security to financial-grade standards.

[0022] 6. By setting up a three-level risk management strategy: low-risk automatic execution (response time <500ms), medium-risk manual review (encrypted push management terminal), and high-risk immediate blocking, we can effectively balance business efficiency and risk control.

[0023] 7. The activation result is encrypted into first-level encrypted data using the RC6 algorithm. The first-level encrypted data is split at a ratio of 1:4 and the order is swapped to obtain second-level encrypted data. The second-level encrypted data is encrypted into an encrypted notification using the Twofish algorithm. At least three security measures are taken (RC6 algorithm, data splitting and order swapping, and Twofish algorithm), greatly improving the security of activation result transmission.

[0024] 8. The real-time activation log is encrypted into the first encrypted data using the IDEA algorithm. A random string of specified length is added to the specified position of the first encrypted data to obtain the second encrypted data. The second encrypted data is encrypted into an encrypted log using the SM4 algorithm. The encrypted log is stored and distributedly backed up. At least four security measures (IDEA algorithm, random string, SM4 algorithm, distributed backup) are taken throughout, greatly improving the security of activation log storage and facilitating later tracing.

[0025] 9. The dataset is divided into training, validation, and test sets in proportion with an 8:1:1 split ratio, which not only ensures the adequacy of training data (80%), but also retains an effective sample size for model tuning (10% validation set) and independent evaluation (10% test set).

[0026] 10. By dynamically adjusting key parameters during model training, including learning rate (to control gradient step size), learning decay rate (to adaptively adjust learning rhythm), random dropout rate (Dropout to prevent overfitting), batch size (to balance training speed and stability), number of iterations (training adequacy), and other core parameters, a multi-dimensional optimization coverage is formed, which greatly improves model performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0028] Figure 1 This is a flow chart of a virtual machine resource provisioning method combined with multimodal risk assessment in the present invention.

[0029] Figure 2 It is a structural diagram of a virtual machine resource provisioning system combined with multimodal risk assessment of the present invention. DETAILED DESCRIPTION

[0030] The technical solution in the embodiments of the present application has the following overall idea: the virtual machine resource activation operation is performed based on the pre-trained order semantic recognition model, risk assessment model and resource dynamic allocation model. When the risk assessment result is low risk, the entire process is automatically operated without human intervention; the order semantic recognition model is used to effectively identify semantic conflicts in the order text, improve the accuracy of configuration requirement identification, and avoid human intervention as much as possible; the resource heat map can effectively integrate resource fragments, and the dynamic allocation of computing resources is combined with the resource dynamic allocation model to effectively utilize computing resources; during the virtual machine resource activation process, the real-time virtual machine purchase order is verified to prevent the order from being tampered with, and the risk assessment of the risk assessment model is combined to improve the efficiency, resource utilization and security of virtual machine resource activation.

[0031] Please refer to Figures 1 to 2 As shown, a preferred embodiment of a virtual machine resource provisioning method combined with multimodal risk assessment of the present invention includes the following steps: Step S1: The cloud platform creates an order semantic recognition model, a risk assessment model, and a resource dynamic allocation model; Step S2: The cloud platform obtains a large amount of historical multimodal data including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs, and constructs a data set after preprocessing and annotating each of the historical multimodal data; Step S3: The cloud platform trains the order semantic recognition model, the risk assessment model, and the resource dynamic allocation model using the data set, and deploys the trained order semantic recognition model, the risk assessment model, and the resource dynamic allocation model; Step S4: The cloud platform collects computing resource data in real time and constructs a resource heat map based on the collected computing resource data; Step S5: The cloud platform obtains a real-time virtual machine purchase order, verifies and parses the real-time virtual machine purchase order to obtain the order content and user account, and obtains real-time multimodal data based on the user account; that is, all data related to the user account is obtained for subsequent risk assessment; Step S6: The cloud platform inputs the order content into the deployed order semantic recognition model to obtain configuration requirements, and inputs the real-time multimodal data into the risk assessment model to obtain risk assessment results; Step S7: The cloud platform invokes the dynamic resource allocation model to allocate computing resources based on the risk assessment results, configuration requirements, and resource heat map, thereby performing a virtual machine resource provisioning operation. Subsequently, the cloud platform can elastically expand or shrink the virtual machine based on the resource heat map and the dynamic resource allocation model to cope with sudden loads and improve resource utilization. Step S8: The cloud platform feeds back the activation result and records the real-time activation log; In step S1, the order semantic recognition model is constructed based on an input representation module, a context encoding module, a multi-task decoding module, and a post-processing enhancement module; the input representation module is used to convert a virtual machine purchase order into an order vector; the context encoding module is constructed based on a BiLSTM network and a Transformer network, the BiLSTM network is used to extract local dependency features from the order vector, and the Transformer network is used to extract global dependency features from the order vector; the multi-task decoding module is used to extract demand parameters, demand strength, and implicit demand from the local dependency features and the global dependency features; the post-processing enhancement module is used to perform Row semantic normalization identifies configuration requirements. The BiLSTM network can capture serialized configuration descriptions (e.g., "must not be less than 4 cores"). The Transformer network can identify cross-sentence associations (e.g., "same configuration as previous order"). During implementation, constraints can also be configured (e.g., "SSD must be equipped with ≥ 4-core CPU"). Requirement parameters include: {Type: CPU, Value: 8 cores}, {Type: Storage, Value: 500GB SSD}. Requirement strength includes hard requirements ("must be equipped with a GPU") and soft requirements ("preferably backup storage"). Implicit requirements, for example, can be identified by automatically associating GPU / TensorCore requirements with "for deep learning"; semantic normalization, for example, standardizes "16GB memory" and "16GB RAM" into a standard format. By setting up an order semantic recognition model based on the input representation module, context encoding module, multi-task decoding module and post-processing enhancement module, combined with hierarchical feature extraction (local→global), multi-task joint optimization (explicit demand parameters and demand intensity + implicit demand), heterogeneous model fusion (BiLSTM+Transformer) and output normalization, deep semantic analysis of virtual machine purchase orders is achieved, thereby greatly improving the accuracy of configuration demand identification.

[0032] The risk assessment model is constructed based on a multimodal input module, a feature extraction module, a feature fusion module, and a risk output module; the multimodal input module is used to input multimodal data; the feature extraction module is used to extract multimodal features including semantic conflict level, abnormal operation frequency, request source geographical distribution, and resource over-allocation ratio from the multimodal data; the feature fusion module is used to fuse the multimodal features through a random forest classifier to obtain a fused feature; the risk output module is used to output a risk assessment result carrying a risk level based on the fused feature; the risk level is low risk, medium risk, or high risk; the semantic conflict level is used to evaluate configuration rationality; the abnormal operation frequency is used to evaluate account credibility; the request source geographical distribution is used to evaluate fraud risk; and the resource over-allocation ratio is used to predict stability; the multimodal features are fused through a random forest classifier, which can be: Risk_Score = 0.3 * semantic risk (semantic conflict level) + 0.2 * behavioral anomaly (abnormal operation frequency) + 0.25 * device suspicion (geographic distribution of request sources) + 0.25 * system load (resource overallocation ratio).

[0033] The feature extraction module of the risk assessment model focuses on four major features: semantic conflict level, abnormal operation frequency, geographical distribution and resource over-allocation ratio. It covers key dimensions such as risk behavior pattern (abnormal frequency), intention recognition (semantic conflict), environmental anomalies (geographic anomalies) and system vulnerability (resource over-allocation). It has clear business explanations. For example, geographical distribution anomalies can identify database stuffing attacks, and resource over-allocation warning system overload risks. It can comprehensively assess the risks in the process of virtual machine resource activation, thereby greatly improving the security of virtual machine resource activation.

[0034] The dynamic resource allocation model is built based on a dynamic graph neural network and is used to generate resource allocation plans based on configuration requirements and resource heat maps.

[0035] By using a dynamic graph neural network to analyze resource heat maps (based on real-time indicators such as CPU / memory / network) in real time, combined with a sliding time window update strategy, flexible resource allocation scheduling is achieved. Compared with static allocation strategies, the utilization rate of computing resources is greatly improved.

[0036] A resource heat map is a visualization tool that uses color to represent resource distribution. It can intuitively demonstrate the density of computing resources in different regions or nodes, providing an important reference for resource allocation. Dynamic graph neural networks are capable of processing dynamically changing graph data. They not only capture the relationships between nodes but also adapt to changes in the graph's topology over time. This makes them suitable for resource allocation scenarios where resource demand and availability change dynamically over time.

[0037] Data from a resource heat map, including information such as resource location, quantity, and demand, is used as input. This data is converted into node features in a dynamic graph neural network, and the connections between nodes are constructed based on resource dependencies or interactions. Dynamic graph neural networks are capable of learning the complex relationships between node features and graph structure. Through training, the network can identify key patterns and patterns in resource allocation, such as which regions have high resource demand but insufficient supply, and which regions have excess resources. Based on these learned patterns and patterns, dynamic graph neural networks can predict optimal resource allocation plans. They can take into account dynamic resource changes and adjust allocation strategies in real time to meet the needs of different regions while optimizing resource utilization.

[0038] Furthermore, the step S2 is specifically as follows: The cloud platform acquires a large amount of historical multimodal data including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs; the historical virtual machine purchase orders include at least the order number, order content, purchase time, purchaser information, and service terms; the historical purchase behavior data includes at least user attributes, order time, purchase frequency, payment method, billing method, IP address, purchase channel, browsing behavior, and interaction behavior; the historical resource data includes at least CPU utilization, memory utilization, storage utilization, storage growth trend, network traffic, network type, network bandwidth, and network topology; the historical activation log includes at least transaction records, activation time, virtual machine name, virtual machine configuration, and user account; Performing preprocessing on each of the historical multimodal data, including at least stop word removal, outlier removal, missing value filling, word segmentation, and format unification; The pre-processed historical multimodal data are labeled with configuration requirements, risk levels, resource heat maps, and resource allocation plans, and a data set is constructed based on the labeled historical multimodal data.

[0039] Furthermore, the step S3 is specifically as follows: The cloud platform divides the dataset into a training set, a validation set, and a test set based on a ratio of 8:1:1. The order semantic recognition model, the risk assessment model, and the dynamic resource allocation model are trained using the corresponding data in the dataset. During the training process, the hyperparameters including at least the learning rate, the learning decay rate, the random dropout rate, the batch size, and the number of iterations are continuously optimized. The weights and biases of the order semantic recognition model, the risk assessment model, and the dynamic resource allocation model are updated in combination with the backpropagation algorithm until the preset convergence conditions are met. Through the split ratio of 8:1:1, the dataset is proportionally allocated to the training set, validation set, and test set, which not only ensures the adequacy of training data (80%), but also retains the effective sample size for model tuning (10% validation set) and independent evaluation (10% test set).

[0040] By dynamically adjusting key parameters during model training, including learning rate (controlling gradient step size), learning decay rate (adaptively adjusting learning rhythm), random dropout rate (Dropout to prevent overfitting), batch size (balancing training speed and stability), and number of iterations (training adequacy), we achieve multi-dimensional optimization coverage, thereby greatly improving model performance.

[0041] The trained order semantic recognition model, risk assessment model and resource dynamic allocation model are verified in turn through the verification set, and the verified order semantic recognition model, risk assessment model and resource dynamic allocation model are tested in turn through the test set, and the tested order semantic recognition model, risk assessment model and resource dynamic allocation model are converted into ONNX format and deployed.

[0042] Furthermore, the step S4 is specifically as follows: The cloud platform collects computing resource data in real time, including at least CPU utilization, memory utilization, storage utilization, storage growth trend, network traffic, network type, network bandwidth, and network topology, performs preprocessing on each computing resource data, including at least cleaning, conversion, and aggregation, maps the preprocessed computing resource data based on a preset display method to obtain a resource heat map, and updates the resource heat map through a preset sliding time window; that is, maps each computing resource data to a node corresponding to the network topology to obtain a resource heat map; To dynamically update the resource heat map, a preset sliding time window technology was introduced. This technology creates training samples using fixed-size windows to capture patterns in time series. During the resource heat map update process, the sliding window moves at preset intervals (such as every second or every minute). Each movement retrieves the latest computing resource data within the window and updates the heat map accordingly. This ensures that the heat map reflects the latest status of computing resources in real time, helping users to keep abreast of resource usage and changing trends.

[0043] The step S5 is specifically as follows: The cloud platform obtains a real-time virtual machine purchase order carrying an encrypted message, a hardware fingerprint, a timestamp, and a hash value; the encrypted message is obtained by encrypting the order content and the user account, and the specific encryption rules are as follows: encrypting the order content and the user account into a first layer of encrypted data using the AES algorithm, converting the first layer of encrypted data into hexadecimal data, swapping the number 6 and the letter A, and the number 9 and the letter B in the hexadecimal data to obtain a second layer of encrypted data, encrypting the second layer of encrypted data into a third layer of encrypted data using the 3DES algorithm, shifting each character of the third layer of encrypted data right by 8 bits to obtain a fourth layer of encrypted data, and encrypting the fourth layer of encrypted data into an encrypted message using the RSA algorithm; the hardware fingerprint is obtained by concatenating the device serial number and the MAC address; the hash value is obtained by hashing the encrypted message, the hardware fingerprint, and the timestamp; By designing a multi-layer hybrid encryption mechanism (AES+3DES+RSA+character replacement) in real-time virtual machine purchase orders, combined with hardware fingerprint (device serial number + MAC address), timestamp, and hash value verification, it can resist man-in-the-middle attacks and replay attacks, and reduce the cost of cracking encrypted messages to the order of 2^128, bringing security to financial-grade standards.

[0044] The cloud platform parses the received real-time virtual machine purchase order to obtain an encrypted message, a hardware fingerprint, a timestamp, and a hash value. After performing an integrity check using the hash value, an aging check is performed using the timestamp, and then a legitimacy check is performed using the hardware fingerprint. The encrypted message is then decrypted into four layers of encrypted data using the RSA algorithm, and each character of the four layers of encrypted data is circularly shifted to the left by 8 bits to obtain three layers of encrypted data. The three layers of encrypted data are decrypted into two layers of encrypted data using the 3DES algorithm, and the number 6 and the letter A in the two layers of encrypted data are swapped, and the number 9 and the letter B are swapped to obtain hexadecimal data. The hexadecimal data is inversely converted to obtain a layer of encrypted data. The layer of encrypted data is decrypted into order content and a user account using the AES algorithm, and real-time multimodal data is obtained based on the user account.

[0045] Furthermore, the step S7 is specifically as follows: The cloud platform analyzes the risk assessment result. When the risk assessment result is low risk, the configuration requirements and the resource heat map are input into the resource dynamic allocation model to obtain a resource allocation plan, and the virtual machine resource provisioning operation is automatically performed based on the resource allocation plan. When the risk assessment result is medium risk, the configuration requirements and the resource heat map are input into the resource dynamic allocation model to obtain a resource allocation plan, and the resource allocation plan is encrypted and pushed to a pre-associated management terminal in real time for approval. When the risk assessment result is high risk, the virtual machine resource provisioning operation is refused. By setting up a three-level risk management strategy: low-risk automatic execution (response time <500ms), medium-risk manual review (encrypted push management terminal), and high-risk immediate blocking, we can effectively balance business efficiency and risk control.

[0046] The step S8 is specifically as follows: The cloud platform generates an activation result indicating activation success or failure, encrypts the activation result into first-level encrypted data using the RC6 algorithm, splits the first-level encrypted data into second-level encrypted data at a ratio of 1:4 and swaps the order of the first and second-level encrypted data, encrypts the second-level encrypted data into an encrypted notification using the Twofish algorithm, and feeds back the encrypted notification. The activation result is encrypted into first-level encrypted data using the RC6 algorithm. The first-level encrypted data is split at a ratio of 1:4 and the order is swapped to obtain second-level encrypted data. The second-level encrypted data is encrypted into an encrypted notification using the Twofish algorithm. At least three security measures are taken (RC6 algorithm, data splitting and order swapping, and Twofish algorithm), greatly improving the security of activation result transmission.

[0047] The cloud platform records real-time activation logs, encrypts the real-time activation logs into first encrypted data using the IDEA algorithm, adds a random string of a specified length to a specified position of the first encrypted data to obtain second encrypted data, encrypts the second encrypted data into an encrypted log using the SM4 algorithm, and stores and distributes the encrypted logs.

[0048] The real-time activation log is encrypted into the first encrypted data using the IDEA algorithm. A random string of specified length is added to the specified position of the first encrypted data to obtain the second encrypted data. The second encrypted data is encrypted into an encrypted log using the SM4 algorithm. The encrypted log is stored and distributedly backed up. At least four security measures (IDEA algorithm, random string, SM4 algorithm, distributed backup) are taken throughout, greatly improving the security of activation log storage and facilitating subsequent tracing.

[0049] A preferred embodiment of a virtual machine resource provisioning system combined with multimodal risk assessment of the present invention includes the following modules: The model creation module is used for the cloud platform to create an order semantic recognition model, a risk assessment model, and a resource dynamic allocation model; A dataset construction module is used for the cloud platform to obtain a large amount of historical multimodal data, including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs, and to construct a dataset after preprocessing and annotating each of the historical multimodal data; A model training module is used for the cloud platform to train the order semantic recognition model, risk assessment model and resource dynamic allocation model through the data set, and deploy the trained order semantic recognition model, risk assessment model and resource dynamic allocation model; A resource heat map construction module is used for collecting computing resource data in real time on the cloud platform and constructing a resource heat map based on the collected computing resource data; The purchase order acquisition module is used by the cloud platform to obtain real-time virtual machine purchase orders, verify and parse the real-time virtual machine purchase orders to obtain order content and user accounts, and obtain real-time multimodal data based on the user accounts; that is, to obtain all data related to the user accounts for subsequent risk assessment; A semantic recognition risk assessment module is used for the cloud platform to input the order content into the deployed order semantic recognition model to obtain configuration requirements, and input the real-time multimodal data into the risk assessment model to obtain risk assessment results; A virtual machine resource provisioning module is used by the cloud platform to call the resource dynamic allocation model to allocate computing resources based on the risk assessment results, configuration requirements, and resource heat map, so as to perform virtual machine resource provisioning operations; and the virtual machine can be elastically expanded or reduced in capacity based on the resource heat map and the resource dynamic allocation model to cope with sudden loads and improve resource utilization; The activation result feedback module is used to feedback the activation results to the cloud platform and record the real-time activation log; In the model creation module, the order semantic recognition model is constructed based on an input representation module, a context encoding module, a multi-task decoding module, and a post-processing enhancement module; the input representation module is used to convert a virtual machine purchase order into an order vector; the context encoding module is constructed based on a BiLSTM network and a Transformer network, the BiLSTM network is used to extract local dependency features from the order vector, and the Transformer network is used to extract global dependency features from the order vector; the multi-task decoding module is used to extract demand parameters, demand intensity, and implicit requirements from the local dependency features and the global dependency features; the post-processing enhancement module is used to perform semantic normalization on the demand parameters, demand intensity, and implicit requirements to obtain configuration requirements; the BiLSTM network can capture serialized configuration descriptions (such as "not less than 4 cores"); the Transformer network can identify cross-sentence associations (such as "same as the previous order configuration"); in specific implementation, constraint rules can also be configured (such as "SSD must be equipped with ≥ 4-core CPU"); demand parameters are for example: {type: CPU, value: 8 cores}, {type: storage, value: 500GB SSD}; requirement strength includes: hard requirements ("must be equipped with a GPU") and soft requirements ("backup storage is best"); implicit requirements, such as identifying "for deep learning" and automatically associating GPU / TensorCore requirements; semantic normalization, such as unifying "16G memory" and "16GB RAM" into a standard format; By setting up an order semantic recognition model based on the input representation module, context encoding module, multi-task decoding module and post-processing enhancement module, combined with hierarchical feature extraction (local→global), multi-task joint optimization (explicit demand parameters and demand intensity + implicit demand), heterogeneous model fusion (BiLSTM+Transformer) and output normalization, deep semantic analysis of virtual machine purchase orders is achieved, thereby greatly improving the accuracy of configuration demand identification.

[0050] The risk assessment model is constructed based on a multimodal input module, a feature extraction module, a feature fusion module, and a risk output module; the multimodal input module is used to input multimodal data; the feature extraction module is used to extract multimodal features including semantic conflict level, abnormal operation frequency, request source geographical distribution, and resource over-allocation ratio from the multimodal data; the feature fusion module is used to fuse the multimodal features through a random forest classifier to obtain a fused feature; the risk output module is used to output a risk assessment result carrying a risk level based on the fused feature; the risk level is low risk, medium risk, or high risk; the semantic conflict level is used to evaluate configuration rationality; the abnormal operation frequency is used to evaluate account credibility; the request source geographical distribution is used to evaluate fraud risk; and the resource over-allocation ratio is used to predict stability; the multimodal features are fused through a random forest classifier, which can be: Risk_Score = 0.3 * semantic risk (semantic conflict level) + 0.2 * behavioral anomaly (abnormal operation frequency) + 0.25 * device suspicion (geographic distribution of request sources) + 0.25 * system load (resource overallocation ratio).

[0051] The feature extraction module of the risk assessment model focuses on four major features: semantic conflict level, abnormal operation frequency, geographical distribution and resource over-allocation ratio. It covers key dimensions such as risk behavior pattern (abnormal frequency), intention recognition (semantic conflict), environmental anomalies (geographic anomalies) and system vulnerability (resource over-allocation). It has clear business explanations. For example, geographical distribution anomalies can identify database stuffing attacks, and resource over-allocation warning system overload risks. It can comprehensively assess the risks in the process of virtual machine resource activation, thereby greatly improving the security of virtual machine resource activation.

[0052] The dynamic resource allocation model is built based on a dynamic graph neural network and is used to generate resource allocation plans based on configuration requirements and resource heat maps.

[0053] By using a dynamic graph neural network to analyze resource heat maps (based on real-time indicators such as CPU / memory / network) in real time, combined with a sliding time window update strategy, flexible resource allocation scheduling is achieved. Compared with static allocation strategies, the utilization rate of computing resources is greatly improved.

[0054] A resource heat map is a visualization tool that uses color to represent resource distribution. It can intuitively demonstrate the density of computing resources in different regions or nodes, providing an important reference for resource allocation. Dynamic graph neural networks are capable of processing dynamically changing graph data. They not only capture the relationships between nodes but also adapt to changes in the graph's topology over time. This makes them suitable for resource allocation scenarios where resource demand and availability change dynamically over time.

[0055] Data from a resource heat map, including information such as resource location, quantity, and demand, is used as input. This data is converted into node features in a dynamic graph neural network, and the connections between nodes are constructed based on resource dependencies or interactions. Dynamic graph neural networks are capable of learning the complex relationships between node features and graph structure. Through training, the network can identify key patterns and patterns in resource allocation, such as which regions have high resource demand but insufficient supply, and which regions have excess resources. Based on these learned patterns and patterns, dynamic graph neural networks can predict optimal resource allocation plans. They can take into account dynamic resource changes and adjust allocation strategies in real time to meet the needs of different regions while optimizing resource utilization.

[0056] The dataset construction module is specifically used for: The cloud platform acquires a large amount of historical multimodal data including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs; the historical virtual machine purchase orders include at least the order number, order content, purchase time, purchaser information, and service terms; the historical purchase behavior data includes at least user attributes, order time, purchase frequency, payment method, billing method, IP address, purchase channel, browsing behavior, and interaction behavior; the historical resource data includes at least CPU utilization, memory utilization, storage utilization, storage growth trend, network traffic, network type, network bandwidth, and network topology; the historical activation log includes at least transaction records, activation time, virtual machine name, virtual machine configuration, and user account; Performing preprocessing on each of the historical multimodal data, including at least stop word removal, outlier removal, missing value filling, word segmentation, and format unification; The pre-processed historical multimodal data are labeled with configuration requirements, risk levels, resource heat maps, and resource allocation plans, and a data set is constructed based on the labeled historical multimodal data.

[0057] The model training module is specifically used for: The cloud platform divides the dataset into a training set, a validation set, and a test set based on a ratio of 8:1:1. The order semantic recognition model, the risk assessment model, and the dynamic resource allocation model are trained using the corresponding data in the dataset. During the training process, the hyperparameters including at least the learning rate, the learning decay rate, the random dropout rate, the batch size, and the number of iterations are continuously optimized. The weights and biases of the order semantic recognition model, the risk assessment model, and the dynamic resource allocation model are updated in combination with the backpropagation algorithm until the preset convergence conditions are met. Through the split ratio of 8:1:1, the dataset is proportionally allocated to the training set, validation set, and test set, which not only ensures the adequacy of training data (80%), but also retains the effective sample size for model tuning (10% validation set) and independent evaluation (10% test set).

[0058] By dynamically adjusting key parameters during model training, including learning rate (controlling gradient step size), learning decay rate (adaptively adjusting learning rhythm), random dropout rate (Dropout to prevent overfitting), batch size (balancing training speed and stability), and number of iterations (training adequacy), we achieve multi-dimensional optimization coverage, thereby greatly improving model performance.

[0059] The trained order semantic recognition model, risk assessment model and resource dynamic allocation model are verified in turn through the verification set, and the verified order semantic recognition model, risk assessment model and resource dynamic allocation model are tested in turn through the test set, and the tested order semantic recognition model, risk assessment model and resource dynamic allocation model are converted into ONNX format and deployed.

[0060] The resource heat map construction module is specifically used to: The cloud platform collects computing resource data in real time, including at least CPU utilization, memory utilization, storage utilization, storage growth trend, network traffic, network type, network bandwidth, and network topology, performs preprocessing on each computing resource data, including at least cleaning, conversion, and aggregation, maps the preprocessed computing resource data based on a preset display method to obtain a resource heat map, and updates the resource heat map through a preset sliding time window; that is, maps each computing resource data to a node corresponding to the network topology to obtain a resource heat map; To dynamically update the resource heat map, a preset sliding time window technology was introduced. This technology creates training samples using fixed-size windows to capture patterns in time series. During the resource heat map update process, the sliding window moves at preset intervals (such as every second or every minute). Each movement retrieves the latest computing resource data within the window and updates the heat map accordingly. This ensures that the heat map reflects the latest status of computing resources in real time, helping users to keep abreast of resource usage and changing trends.

[0061] The purchase order acquisition module is specifically used to: The cloud platform obtains a real-time virtual machine purchase order carrying an encrypted message, a hardware fingerprint, a timestamp, and a hash value; the encrypted message is obtained by encrypting the order content and the user account, and the specific encryption rules are as follows: encrypting the order content and the user account into a first layer of encrypted data using the AES algorithm, converting the first layer of encrypted data into hexadecimal data, swapping the number 6 and the letter A, and the number 9 and the letter B in the hexadecimal data to obtain a second layer of encrypted data, encrypting the second layer of encrypted data into a third layer of encrypted data using the 3DES algorithm, shifting each character of the third layer of encrypted data right by 8 bits to obtain a fourth layer of encrypted data, and encrypting the fourth layer of encrypted data into an encrypted message using the RSA algorithm; the hardware fingerprint is obtained by concatenating the device serial number and the MAC address; the hash value is obtained by hashing the encrypted message, the hardware fingerprint, and the timestamp; By designing a multi-layer hybrid encryption mechanism (AES+3DES+RSA+character replacement) in real-time virtual machine purchase orders, combined with hardware fingerprint (device serial number + MAC address), timestamp, and hash value verification, it can resist man-in-the-middle attacks and replay attacks, and reduce the cost of cracking encrypted messages to the order of 2^128, bringing security to financial-grade standards.

[0062] The cloud platform parses the received real-time virtual machine purchase order to obtain an encrypted message, a hardware fingerprint, a timestamp, and a hash value. After performing an integrity check using the hash value, an aging check is performed using the timestamp, and then a legitimacy check is performed using the hardware fingerprint. The encrypted message is then decrypted into four layers of encrypted data using the RSA algorithm, and each character of the four layers of encrypted data is circularly shifted to the left by 8 bits to obtain three layers of encrypted data. The three layers of encrypted data are decrypted into two layers of encrypted data using the 3DES algorithm, and the number 6 and the letter A in the two layers of encrypted data are swapped, and the number 9 and the letter B are swapped to obtain hexadecimal data. The hexadecimal data is inversely converted to obtain a layer of encrypted data. The layer of encrypted data is decrypted into order content and a user account using the AES algorithm, and real-time multimodal data is obtained based on the user account.

[0063] The virtual machine resource provisioning module is specifically used to: The cloud platform analyzes the risk assessment result. When the risk assessment result is low risk, the configuration requirements and the resource heat map are input into the resource dynamic allocation model to obtain a resource allocation plan, and the virtual machine resource provisioning operation is automatically performed based on the resource allocation plan. When the risk assessment result is medium risk, the configuration requirements and the resource heat map are input into the resource dynamic allocation model to obtain a resource allocation plan, and the resource allocation plan is encrypted and pushed to a pre-associated management terminal in real time for approval. When the risk assessment result is high risk, the virtual machine resource provisioning operation is refused. By setting up a three-level risk management strategy: low-risk automatic execution (response time <500ms), medium-risk manual review (encrypted push management terminal), and high-risk immediate blocking, we can effectively balance business efficiency and risk control.

[0064] The activation result feedback module is specifically used to: The cloud platform generates an activation result indicating activation success or failure, encrypts the activation result into first-level encrypted data using the RC6 algorithm, splits the first-level encrypted data into second-level encrypted data at a ratio of 1:4 and swaps the order of the first and second-level encrypted data, encrypts the second-level encrypted data into an encrypted notification using the Twofish algorithm, and feeds back the encrypted notification. The activation result is encrypted into first-level encrypted data using the RC6 algorithm. The first-level encrypted data is split at a ratio of 1:4 and the order is swapped to obtain second-level encrypted data. The second-level encrypted data is encrypted into an encrypted notification using the Twofish algorithm. At least three security measures are taken (RC6 algorithm, data splitting and order swapping, and Twofish algorithm), greatly improving the security of activation result transmission.

[0065] The cloud platform records real-time activation logs, encrypts the real-time activation logs into first encrypted data using the IDEA algorithm, adds a random string of a specified length to a specified position of the first encrypted data to obtain second encrypted data, encrypts the second encrypted data into an encrypted log using the SM4 algorithm, and stores and distributes the encrypted logs.

[0066] The real-time activation log is encrypted into the first encrypted data using the IDEA algorithm. A random string of specified length is added to the specified position of the first encrypted data to obtain the second encrypted data. The second encrypted data is encrypted into an encrypted log using the SM4 algorithm. The encrypted log is stored and distributedly backed up. At least four security measures (IDEA algorithm, random string, SM4 algorithm, distributed backup) are taken throughout, greatly improving the security of activation log storage and facilitating subsequent tracing.

[0067] In summary, the advantages of the present invention are: 1. Create an order semantic recognition model, a risk assessment model, and a resource dynamic allocation model through the cloud platform, obtain a large amount of historical multimodal data including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs, pre-process and annotate each historical multimodal data to build a data set, and use the data set to train and deploy the order semantic recognition model, risk assessment model, and resource dynamic allocation model; then the cloud platform collects computing resource data in real time to build a resource heat map, verifies and parses the obtained real-time virtual machine purchase orders to obtain order content and user accounts, obtains real-time multimodal data based on the user account, inputs the order content into the deployed order semantic recognition model to obtain configuration requirements, and inputs the real-time multimodal data into the risk assessment model to obtain risk assessment results; then, based on the risk assessment results, configuration requirements, and resource heat map, calls resources. The dynamic allocation model executes the allocation of computing resources to execute the virtual machine resource activation operation. The cloud platform feedbacks the activation results and records the real-time activation log. That is, the virtual machine resource activation operation is executed based on the pre-trained order semantic recognition model, risk assessment model and resource dynamic allocation model. When the risk assessment result is low risk, the entire process is automatic and no human intervention is required. The order semantic recognition model can effectively identify semantic conflicts in order texts, improve the accuracy of configuration requirement identification, and avoid human intervention as much as possible. The resource heat map can effectively integrate resource fragments, and combine with the resource dynamic allocation model to dynamically allocate computing resources, which can effectively utilize computing resources. In addition, during the virtual machine resource activation process, the real-time virtual machine purchase order is verified to prevent the order from being tampered with. Combined with the risk assessment of the risk assessment model, the efficiency, resource utilization and security of virtual machine resource activation are ultimately greatly improved.

[0068] 2. By setting up an order semantic recognition model based on the input representation module, context encoding module, multi-task decoding module and post-processing enhancement module, combined with hierarchical feature extraction (local to global), multi-task joint optimization (explicit demand parameters and demand intensity + implicit demand), heterogeneous model fusion (BiLSTM + Transformer) and output normalization, deep semantic analysis of virtual machine purchase orders is achieved, thereby greatly improving the accuracy of configuration demand identification.

[0069] 3. The feature extraction module of the risk assessment model focuses on four major features: semantic conflict level, abnormal operation frequency, geographical distribution, and resource over-allocation ratio. It covers key dimensions such as risk behavior pattern (abnormal frequency), intent recognition (semantic conflict), environmental anomalies (geographic anomalies), and system vulnerability (resource over-allocation). It has clear business explanations. For example, geographical distribution anomalies can identify database stuffing attacks, and resource over-allocation can warn of system overload risks. It can comprehensively assess the risks in the process of virtual machine resource activation, thereby greatly improving the security of virtual machine resource activation.

[0070] 4. By using a dynamic graph neural network to analyze resource heat maps in real time (based on real-time indicators such as CPU, memory, and network), combined with a sliding time window update strategy, flexible resource allocation scheduling is achieved. Compared with static allocation strategies, the utilization rate of computing resources is greatly improved.

[0071] 5. By designing a multi-layer hybrid encryption mechanism (AES+3DES+RSA+character replacement) in real-time virtual machine purchase orders, combined with hardware fingerprint (device serial number + MAC address), timestamp, and hash value verification, it can resist man-in-the-middle attacks and replay attacks, and reduce the cost of cracking encrypted messages to the order of 2^128, bringing security to financial-grade standards.

[0072] 6. By setting up a three-level risk management strategy: low-risk automatic execution (response time <500ms), medium-risk manual review (encrypted push management terminal), and high-risk immediate blocking, we can effectively balance business efficiency and risk control.

[0073] 7. The activation result is encrypted into first-level encrypted data using the RC6 algorithm. The first-level encrypted data is split at a ratio of 1:4 and the order is swapped to obtain second-level encrypted data. The second-level encrypted data is encrypted into an encrypted notification using the Twofish algorithm. At least three security measures are taken (RC6 algorithm, data splitting and order swapping, and Twofish algorithm), greatly improving the security of activation result transmission.

[0074] 8. The real-time activation log is encrypted into the first encrypted data using the IDEA algorithm. A random string of specified length is added to the specified position of the first encrypted data to obtain the second encrypted data. The second encrypted data is encrypted into an encrypted log using the SM4 algorithm. The encrypted log is stored and distributedly backed up. At least four security measures (IDEA algorithm, random string, SM4 algorithm, distributed backup) are taken throughout, greatly improving the security of activation log storage and facilitating later tracing.

[0075] 9. The dataset is divided into training, validation, and test sets in proportion with an 8:1:1 split ratio, which not only ensures the adequacy of training data (80%), but also retains an effective sample size for model tuning (10% validation set) and independent evaluation (10% test set).

[0076] 10. By dynamically adjusting key parameters during model training, including learning rate (to control gradient step size), learning decay rate (to adaptively adjust learning rhythm), random dropout rate (Dropout to prevent overfitting), batch size (to balance training speed and stability), number of iterations (training adequacy), and other core parameters, a multi-dimensional optimization coverage is formed, which greatly improves model performance.

[0077] Although the specific embodiments of the present invention are described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for provisioning virtual machine resources in combination with multimodal risk assessment, characterized by: The steps include: Step S1: The cloud platform creates an order semantic recognition model, a risk assessment model, and a resource dynamic allocation model; Step S2: The cloud platform obtains a large amount of historical multimodal data including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs, and constructs a data set after preprocessing and annotating each of the historical multimodal data; Step S3: The cloud platform trains the order semantic recognition model, the risk assessment model, and the resource dynamic allocation model using the data set, and deploys the trained order semantic recognition model, the risk assessment model, and the resource dynamic allocation model; Step S4: The cloud platform collects computing resource data in real time and constructs a resource heat map based on the collected computing resource data; Step S5: The cloud platform obtains a real-time virtual machine purchase order, verifies and parses the real-time virtual machine purchase order to obtain order content and a user account, and obtains real-time multimodal data based on the user account; Step S6: The cloud platform inputs the order content into the deployed order semantic recognition model to obtain configuration requirements, and inputs the real-time multimodal data into the risk assessment model to obtain risk assessment results; Step S7: The cloud platform calls the resource dynamic allocation model to allocate computing resources based on the risk assessment results, configuration requirements, and resource heat map, so as to perform a virtual machine resource provisioning operation; Step S8: The cloud platform feeds back the activation result and records the real-time activation log.

2. The method for provisioning virtual machine resources in combination with multimodal risk assessment according to claim 1, characterized in that: In step S1, the order semantic recognition model is constructed based on an input representation module, a context encoding module, a multi-task decoding module, and a post-processing enhancement module; the input representation module is used to convert a virtual machine purchase order into an order vector; the context encoding module is constructed based on a BiLSTM network and a Transformer network, the BiLSTM network is used to extract local dependency features from the order vector, and the Transformer network is used to extract global dependency features from the order vector; the multi-task decoding module is used to extract local dependency features and global dependency features from the order vector. Extracting demand parameters, demand strength and implicit demand; the post-processing enhancement module is used to perform semantic normalization on the demand parameters, demand strength and implicit demand to obtain configuration requirements; The risk assessment model is constructed based on a multimodal input module, a feature extraction module, a feature fusion module, and a risk output module; the multimodal input module is used to input multimodal data; the feature extraction module is used to extract multimodal features including semantic conflict level, abnormal operation frequency, geographical distribution of request sources, and resource over-allocation ratio from the multimodal data; the feature fusion module is used to fuse the multimodal features through a random forest classifier to obtain a fused feature; the risk output module is used to output a risk assessment result with a risk level based on the fused feature; the risk level is low risk, medium risk, or high risk; The dynamic resource allocation model is built based on a dynamic graph neural network and is used to generate resource allocation plans based on configuration requirements and resource heat maps; The step S2 is specifically as follows: The cloud platform acquires a large amount of historical multimodal data including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs; the historical virtual machine purchase orders include at least the order number, order content, purchase time, purchaser information, and service terms; the historical purchase behavior data includes at least user attributes, order time, purchase frequency, payment method, billing method, IP address, purchase channel, browsing behavior, and interaction behavior; the historical resource data includes at least CPU utilization, memory utilization, storage utilization, storage growth trend, network traffic, network type, network bandwidth, and network topology; the historical activation log includes at least transaction records, activation time, virtual machine name, virtual machine configuration, and user account; Performing preprocessing on each of the historical multimodal data, including at least stop word removal, outlier removal, missing value filling, word segmentation, and format unification; The pre-processed historical multimodal data are labeled with configuration requirements, risk levels, resource heat maps, and resource allocation plans, and a data set is constructed based on the labeled historical multimodal data.

3. The method for provisioning virtual machine resources in combination with multimodal risk assessment according to claim 1, characterized in that: The step S3 is specifically as follows: The cloud platform divides the dataset into a training set, a validation set, and a test set based on a ratio of 8:1:

1. The order semantic recognition model, the risk assessment model, and the dynamic resource allocation model are trained using the corresponding data in the dataset. During the training process, the hyperparameters including at least the learning rate, the learning decay rate, the random dropout rate, the batch size, and the number of iterations are continuously optimized. The weights and biases of the order semantic recognition model, the risk assessment model, and the dynamic resource allocation model are updated in combination with the backpropagation algorithm until the preset convergence conditions are met. The trained order semantic recognition model, risk assessment model and resource dynamic allocation model are verified in turn through the verification set, and the verified order semantic recognition model, risk assessment model and resource dynamic allocation model are tested in turn through the test set, and the tested order semantic recognition model, risk assessment model and resource dynamic allocation model are converted into ONNX format and deployed.

4. The method for provisioning virtual machine resources in combination with multimodal risk assessment according to claim 1, wherein: The step S4 is specifically as follows: The cloud platform collects computing resource data in real time, including at least CPU utilization, memory utilization, storage utilization, storage growth trend, network traffic, network type, network bandwidth, and network topology; performs preprocessing on each computing resource data, including at least cleaning, conversion, and aggregation; maps the preprocessed computing resource data based on a preset display method to obtain a resource heat map; and updates the resource heat map through a preset sliding time window; The step S5 is specifically as follows: The cloud platform obtains a real-time virtual machine purchase order carrying an encrypted message, a hardware fingerprint, a timestamp, and a hash value; the encrypted message is obtained by encrypting the order content and the user account, and the specific encryption rules are as follows: encrypting the order content and the user account into a first layer of encrypted data using the AES algorithm, converting the first layer of encrypted data into hexadecimal data, swapping the number 6 and the letter A, and the number 9 and the letter B in the hexadecimal data to obtain a second layer of encrypted data, encrypting the second layer of encrypted data into a third layer of encrypted data using the 3DES algorithm, shifting each character of the third layer of encrypted data right by 8 bits to obtain a fourth layer of encrypted data, and encrypting the fourth layer of encrypted data into an encrypted message using the RSA algorithm; the hardware fingerprint is obtained by concatenating the device serial number and the MAC address; the hash value is obtained by hashing the encrypted message, the hardware fingerprint, and the timestamp; The cloud platform parses the received real-time virtual machine purchase order to obtain an encrypted message, a hardware fingerprint, a timestamp, and a hash value. After performing an integrity check using the hash value, an aging check is performed using the timestamp, and then a legitimacy check is performed using the hardware fingerprint. The encrypted message is then decrypted into four layers of encrypted data using the RSA algorithm, and each character of the four layers of encrypted data is circularly shifted to the left by 8 bits to obtain three layers of encrypted data. The three layers of encrypted data are decrypted into two layers of encrypted data using the 3DES algorithm, and the number 6 and the letter A in the two layers of encrypted data are swapped, and the number 9 and the letter B are swapped to obtain hexadecimal data. The hexadecimal data is inversely converted to obtain a layer of encrypted data. The layer of encrypted data is decrypted into order content and a user account using the AES algorithm, and real-time multimodal data is obtained based on the user account.

5. The method for provisioning virtual machine resources in combination with multimodal risk assessment according to claim 1, characterized in that: The step S7 is specifically as follows: The cloud platform analyzes the risk assessment result. When the risk assessment result is low risk, the configuration requirements and the resource heat map are input into the resource dynamic allocation model to obtain a resource allocation plan, and the virtual machine resource provisioning operation is automatically performed based on the resource allocation plan. When the risk assessment result is medium risk, the configuration requirements and the resource heat map are input into the resource dynamic allocation model to obtain a resource allocation plan, and the resource allocation plan is encrypted and pushed to a pre-associated management terminal in real time for approval. When the risk assessment result is high risk, the virtual machine resource provisioning operation is refused. The step S8 is specifically as follows: The cloud platform generates an activation result indicating activation success or failure, encrypts the activation result into first-level encrypted data using the RC6 algorithm, splits the first-level encrypted data into second-level encrypted data at a ratio of 1:4 and swaps the order of the first and second-level encrypted data, encrypts the second-level encrypted data into an encrypted notification using the Twofish algorithm, and feeds back the encrypted notification. The cloud platform records real-time activation logs, encrypts the real-time activation logs into first encrypted data using the IDEA algorithm, adds a random string of a specified length to a specified position of the first encrypted data to obtain second encrypted data, encrypts the second encrypted data into an encrypted log using the SM4 algorithm, and stores and distributes the encrypted logs.

6. A virtual machine resource provisioning system combined with multimodal risk assessment, characterized by: Includes the following modules: The model creation module is used for the cloud platform to create an order semantic recognition model, a risk assessment model, and a resource dynamic allocation model; A dataset construction module is used for the cloud platform to obtain a large amount of historical multimodal data, including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs, and to construct a dataset after preprocessing and annotating each of the historical multimodal data; A model training module is used for the cloud platform to train the order semantic recognition model, risk assessment model and resource dynamic allocation model through the data set, and deploy the trained order semantic recognition model, risk assessment model and resource dynamic allocation model; A resource heat map construction module is used for collecting computing resource data in real time on the cloud platform and constructing a resource heat map based on the collected computing resource data; A purchase order acquisition module is used for the cloud platform to obtain real-time virtual machine purchase orders, verify and parse the real-time virtual machine purchase orders to obtain order content and user accounts, and obtain real-time multimodal data based on the user accounts; A semantic recognition risk assessment module is used for the cloud platform to input the order content into the deployed order semantic recognition model to obtain configuration requirements, and input the real-time multimodal data into the risk assessment model to obtain risk assessment results; A virtual machine resource provisioning module is used for the cloud platform to call the resource dynamic allocation model to allocate computing resources based on the risk assessment results, configuration requirements and resource heat map, so as to perform virtual machine resource provisioning operations; The activation result feedback module is used to feedback the activation results to the cloud platform and record real-time activation logs.

7. The virtual machine resource provisioning system combined with multimodal risk assessment according to claim 6, characterized in that: In the model creation module, the order semantic recognition model is constructed based on the input representation module, the context encoding module, the multi-task decoding module and the post-processing enhancement module; the input representation module is used to convert the virtual machine purchase order into an order vector; the context encoding module is constructed based on the BiLSTM network and the Transformer network, the BiLSTM network is used to extract local dependency features from the order vector, and the Transformer network is used to extract global dependency features from the order vector; the multi-task decoding module is used to extract local dependency features and global dependency features from the order vector. Extracting demand parameters, demand strength and implicit demand; the post-processing enhancement module is used to perform semantic normalization on the demand parameters, demand strength and implicit demand to obtain configuration requirements; The risk assessment model is constructed based on a multimodal input module, a feature extraction module, a feature fusion module, and a risk output module; the multimodal input module is used to input multimodal data; the feature extraction module is used to extract multimodal features including semantic conflict level, abnormal operation frequency, geographical distribution of request sources, and resource over-allocation ratio from the multimodal data; the feature fusion module is used to fuse the multimodal features through a random forest classifier to obtain a fused feature; the risk output module is used to output a risk assessment result with a risk level based on the fused feature; the risk level is low risk, medium risk, or high risk; The dynamic resource allocation model is built based on a dynamic graph neural network and is used to generate resource allocation plans based on configuration requirements and resource heat maps; The dataset construction module is specifically used for: The cloud platform acquires a large amount of historical multimodal data including historical virtual machine purchase orders, historical purchase behavior data, historical resource data, and historical activation logs; the historical virtual machine purchase orders include at least the order number, order content, purchase time, purchaser information, and service terms; the historical purchase behavior data includes at least user attributes, order time, purchase frequency, payment method, billing method, IP address, purchase channel, browsing behavior, and interaction behavior; the historical resource data includes at least CPU utilization, memory utilization, storage utilization, storage growth trend, network traffic, network type, network bandwidth, and network topology; the historical activation log includes at least transaction records, activation time, virtual machine name, virtual machine configuration, and user account; Performing preprocessing on each of the historical multimodal data, including at least stop word removal, outlier removal, missing value filling, word segmentation, and format unification; The pre-processed historical multimodal data are labeled with configuration requirements, risk levels, resource heat maps, and resource allocation plans, and a data set is constructed based on the labeled historical multimodal data.

8. The virtual machine resource provisioning system combined with multimodal risk assessment according to claim 6, characterized in that: The model training module is specifically used for: The cloud platform divides the dataset into a training set, a validation set, and a test set based on a ratio of 8:1:

1. The order semantic recognition model, the risk assessment model, and the dynamic resource allocation model are trained using the corresponding data in the dataset. During the training process, the hyperparameters including at least the learning rate, the learning decay rate, the random dropout rate, the batch size, and the number of iterations are continuously optimized. The weights and biases of the order semantic recognition model, the risk assessment model, and the dynamic resource allocation model are updated in combination with the backpropagation algorithm until the preset convergence conditions are met. The trained order semantic recognition model, risk assessment model and resource dynamic allocation model are verified in turn through the verification set, and the verified order semantic recognition model, risk assessment model and resource dynamic allocation model are tested in turn through the test set, and the tested order semantic recognition model, risk assessment model and resource dynamic allocation model are converted into ONNX format and deployed.

9. The virtual machine resource provisioning system combined with multimodal risk assessment according to claim 6, characterized in that: The resource heat map construction module is specifically used to: The cloud platform collects computing resource data in real time, including at least CPU utilization, memory utilization, storage utilization, storage growth trend, network traffic, network type, network bandwidth, and network topology; performs preprocessing on each computing resource data, including at least cleaning, conversion, and aggregation; maps the preprocessed computing resource data based on a preset display method to obtain a resource heat map; and updates the resource heat map through a preset sliding time window; The purchase order acquisition module is specifically used to: The cloud platform obtains a real-time virtual machine purchase order carrying an encrypted message, a hardware fingerprint, a timestamp, and a hash value; the encrypted message is obtained by encrypting the order content and the user account, and the specific encryption rules are as follows: encrypting the order content and the user account into a first layer of encrypted data using the AES algorithm, converting the first layer of encrypted data into hexadecimal data, swapping the number 6 and the letter A, and the number 9 and the letter B in the hexadecimal data to obtain a second layer of encrypted data, encrypting the second layer of encrypted data into a third layer of encrypted data using the 3DES algorithm, shifting each character of the third layer of encrypted data right by 8 bits to obtain a fourth layer of encrypted data, and encrypting the fourth layer of encrypted data into an encrypted message using the RSA algorithm; the hardware fingerprint is obtained by concatenating the device serial number and the MAC address; the hash value is obtained by hashing the encrypted message, the hardware fingerprint, and the timestamp; The cloud platform parses the received real-time virtual machine purchase order to obtain an encrypted message, a hardware fingerprint, a timestamp, and a hash value. After performing an integrity check using the hash value, an aging check is performed using the timestamp, and then a legitimacy check is performed using the hardware fingerprint. The encrypted message is then decrypted into four layers of encrypted data using the RSA algorithm, and each character of the four layers of encrypted data is circularly shifted to the left by 8 bits to obtain three layers of encrypted data. The three layers of encrypted data are decrypted into two layers of encrypted data using the 3DES algorithm, and the number 6 and the letter A in the two layers of encrypted data are swapped, and the number 9 and the letter B are swapped to obtain hexadecimal data. The hexadecimal data is inversely converted to obtain a layer of encrypted data. The layer of encrypted data is decrypted into order content and a user account using the AES algorithm, and real-time multimodal data is obtained based on the user account.

10. The virtual machine resource provisioning system combined with multimodal risk assessment according to claim 6, characterized in that: The virtual machine resource provisioning module is specifically used to: The cloud platform analyzes the risk assessment result. When the risk assessment result is low risk, the configuration requirements and the resource heat map are input into the resource dynamic allocation model to obtain a resource allocation plan, and the virtual machine resource provisioning operation is automatically performed based on the resource allocation plan. When the risk assessment result is medium risk, the configuration requirements and the resource heat map are input into the resource dynamic allocation model to obtain a resource allocation plan, and the resource allocation plan is encrypted and pushed to a pre-associated management terminal in real time for approval. When the risk assessment result is high risk, the virtual machine resource provisioning operation is refused. The activation result feedback module is specifically used to: The cloud platform generates an activation result indicating activation success or failure, encrypts the activation result into first-level encrypted data using the RC6 algorithm, splits the first-level encrypted data into second-level encrypted data at a ratio of 1:4 and swaps the order of the first and second-level encrypted data, encrypts the second-level encrypted data into an encrypted notification using the Twofish algorithm, and feeds back the encrypted notification. The cloud platform records real-time activation logs, encrypts the real-time activation logs into first encrypted data using the IDEA algorithm, adds a random string of a specified length to a specified position of the first encrypted data to obtain second encrypted data, encrypts the second encrypted data into an encrypted log using the SM4 algorithm, and stores and distributes the encrypted logs.

Images