A cloud terminal intelligent service optimization method and system based on big data analysis

Abstract

The application discloses a cloud terminal intelligent service optimization method and system based on big data analysis, comprising the following steps: S1, collecting the operation log, behavior data and communication data of the cloud terminal, and completing unified preprocessing to generate structured input; S2, inputting the structured data into a time domain double-rotation nested network model, extracting periodic and mutational behavior characteristics, and generating a collaborative feature vector; S3, performing a general Hessian mapping projection optimization based on the collaborative feature vector, and adjusting the model nesting depth and channel configuration; S4, generating a resource scheduling strategy according to the output of the optimized model, including a service instance mapping relationship and a scheduling priority; S5, deploying the resource scheduling strategy to a cloud service control module, and performing task scheduling and service migration; S6, collecting service response and resource utilization feedback, updating model parameters, and realizing scheduling closed-loop optimization. The application realizes cloud terminal service response speedup and resource scheduling intelligentization, and significantly improves system stability and operation efficiency.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and cloud computing technology, and in particular to a method and system for optimizing intelligent services for cloud terminals based on big data analysis. Background Technology

[0002] In existing technologies, with the development of cloud computing and terminal intelligence, cloud service platforms are gradually undertaking a large number of terminal requests, resource scheduling, and application response tasks. As edge access nodes, cloud terminals are responsible for front-end data perception and interaction, undertaking important functions such as service invocation, task execution, and feedback transmission. With the continuous increase in terminal access density and service diversification, traditional service scheduling methods based on static rules or experience weights are no longer sufficient to meet the needs of dynamic, high-frequency, and resource-sensitive application scenarios. At times, in managing complex user behaviors, large-scale heterogeneous resources, and high-concurrency requests, the contradiction between service response efficiency and resource utilization becomes particularly prominent.

[0003] Currently, some technical solutions attempt to introduce machine learning or rule engines for auxiliary optimization, but most methods have limitations in data representation and model structure. First, the lack of uniformity in data preprocessing makes it difficult to form consistent feature representations for log, communication, and behavioral data in terms of dimensional fusion and time-series processing. Second, most existing feature extraction methods rely on single-channel static models, making it difficult to capture the complex dynamic features of periodic behavior and sudden changes in terminal operation, and lacking the ability to model multi-scale semantics in the time domain. Third, existing resource scheduling strategies generally adopt heuristic or rule-driven approaches, failing to jointly model and optimize the relationship between service instance behavior changes and model structure response, resulting in a lack of adaptability, accuracy, and responsiveness in the scheduling strategy. Fourth, the lack of a closed-loop self-optimization mechanism means that feedback data during service scheduling cannot effectively participate in model updates, making it almost impossible to achieve adaptive updates of the strategy as the operational state evolves.

[0004] More importantly, in existing technological systems, deep learning models often operate with a fixed structure. Parameter settings such as the number of channels and nesting levels cannot be restructured based on the dynamic load status and behavioral differences of cloud terminals. The decoupling between model structure and task scheduling is severe, leading to problems such as redundant resource allocation, unstable service response, and untimely instance switching. In addition to these issues, most solutions lack the ability to generate strategies based on structural optimization and behavioral characteristics. They cannot support service scheduling strategies for cloud terminals at the model structure level, which limits the scalability and scheduling intelligence of cloud service systems in high-density scenarios.

[0005] Therefore, there is an urgent need for a cloud terminal intelligent service optimization method that has behavioral perception, multi-scale feature expression capabilities, and can integrate structural self-optimization and strategy self-generation mechanisms. This method can extract temporal collaborative features from big data, dynamically adjust the model structure, and generate high-precision, deployable scheduling strategies based on the optimization results, thereby improving terminal response performance and resource allocation efficiency.

[0006] Therefore, how to provide a cloud terminal intelligent service optimization method and system based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of this invention is to propose a cloud terminal intelligent service optimization method and system based on big data analysis. This invention fully integrates the temporal double-rotation nested network model and the generalized Hessian graph projection optimization algorithm, and describes in detail the feature extraction, structure optimization and strategy generation process for cloud terminal running data. It has the advantages of strong model adaptability, high resource scheduling accuracy and good service response stability.

[0008] A cloud terminal intelligent service optimization method and system based on big data analysis according to an embodiment of the present invention includes the following steps:

[0009] S1. Collect data from cloud terminals, perform unified preprocessing, and generate a structured input dataset;

[0010] S2. Input the structured input dataset into the established temporal double-rotation nested network model. The temporal double-rotation nested network model includes the main rotation channel, the inverse rotation channel, and nested residual blocks, and generates collaborative feature vectors.

[0011] S3. Construct a graph structure, perform generalized Hessian graph projection optimization on the collaborative feature vectors, calculate the second-order gradient response, optimize the nesting depth and channel configuration inside the temporal double-rotation nested network model, and obtain the structurally optimized model;

[0012] S4. Generate a cloud terminal resource scheduling strategy based on the optimized model. The cloud terminal resource scheduling strategy includes service instance mapping relationship and scheduling priority.

[0013] S5. Deploy resource scheduling strategies to the cloud service control module to control cloud terminals to perform task allocation and load migration operations;

[0014] S6. Real-time collection of response time, resource utilization and service consistency results of cloud terminal devices during the execution of the resource scheduling strategy, updating the structured and optimized model to achieve adaptive adjustment of strategy and structure.

[0015] Optionally, the unified preprocessing in step S1 includes performing timestamp alignment, missing value imputation, format standardization, and outlier removal operations on the data collected from the cloud terminal.

[0016] Optionally, S2 specifically includes:

[0017] S21. Represent the structured input dataset as a matrix. The input is fed into the main rotation channel and processed by the rotating convolution kernel. Extract periodic features to obtain feature sequences , where the symbol This indicates a rotational convolution operation;

[0018] S22. Transform the structured input dataset Simultaneously input to the inverse rotation channel, through the inverse rotation convolution kernel Extract mutation features and obtain feature sequences ;

[0019] S23, The feature sequence and The features are merged into a single input feature set and fed into a multi-layered nested residual block structure. Each residual block contains information branch channels and feature conservation gating structures. Cross-scale feature fusion and residual connection operations are performed, and the fused collaborative feature vector is output. .

[0020] Optionally, the cross-scale feature fusion and residual connection operation in S23 specifically includes:

[0021] S231, merge the feature sequences and The synchronous input is fed into a multi-layer nested residual block structure, which contains several scale channel groups. Different channel groups extract feature information of the corresponding scale according to their respective set convolution receptive field size, and generate several sets of scale feature tensors.

[0022] S232. Set up an independent information branch channel for each set of scale feature tensors, and introduce a gating mechanism in each information branch channel to dynamically control whether the output of the feature tensor is passed according to the response characteristics of the current feature tensor.

[0023] S233. Fuse all the scale feature tensors after gating and adjustment, establish residual connections with the input features received by each scale channel group, and construct a nested residual structure with cross-scale fusion path.

[0024] S234. After fusion, output a unified collaborative feature vector. The collaborative feature vector As a synergistic feature representing periodic and mutable behavior.

[0025] Optionally, S3 specifically includes:

[0026] S31, Combine the feature vectors Input is sent to the structure graph module to construct a graph structure corresponding to the model structure. ,in, This represents the set of structural nodes in a neural network. Indicates the connection relationship between nodes;

[0027] S32, in the diagram structure Above, based on collaborative feature vectors The forward propagation path is guided by which the total loss function of the model in the service response and resource prediction tasks is calculated. The second-order gradient response of each structural node to the total loss function is extracted, and the corresponding Hessian matrix is ​​constructed. ;

[0028] S33, Based on the Hessian matrix Local sensitivity changes in the data allow for the selection of the set of structural nodes to be adjusted. The structural update mapping relationship is generated by combining structural constraints.

[0029] S34. Apply the structure update mapping relationship to the graph structure. By adjusting the nesting depth and channel configuration within the temporal double-rotation nested network model through graph projection, the model with optimized structure is output.

[0030] Optionally, S34, which adjusts the nesting depth and channel configuration within the temporal double-rotation nested network model via graph projection, specifically includes:

[0031] S341, In the diagram structure The set of nodes associated with the nested residual structure is identified in the middle. Each node represents a nested residual block, and the node attributes include the corresponding nesting level index and the number of channels.

[0032] S342, Based on Hessian matrix The response strength of a node in a set of nodes The nodes in the algorithm are sorted by their responses, and a subset of nodes with sensitivity below a preset value is selected. Used for structural adjustment;

[0033] S343, Targeting a subset of nodes Based on the preset minimum depth threshold and channel lower limit, the number of nested layers is pruned, or the channel configuration parameters are compressed and adjusted to generate a structural adjustment map.

[0034] S344. Project the structural adjustment diagram back to the original structural diagram. This forms the updated graph structure. The graph structure The corresponding temporal double-rotation nested network model is the structurally optimized model after adjusting the nesting depth and channel configuration.

[0035] Optionally, S4 specifically includes:

[0036] Cloud terminal resource scheduling strategy based on collaborative feature vector The optimized model output and real-time collected response data generation include the following scheduling rules:

[0037] If the collaborative feature vector If the periodic characteristic value represented by the main rotating channel increases in several consecutive time segments, and the channel configuration corresponding to the service instance in the optimized model has reached the current node's resource utilization limit, then the service instance will be migrated to another node, where the channel utilization rate of the other node is lower than the average level.

[0038] If the collaborative feature vector If the non-periodic features represented by the inverse rotation channel suddenly increase within a certain period of time, and the nesting depth of the instance is compressed to the minimum value in the output of the optimized model, then the scheduling frequency of the service instance is reduced, and the occupied part of the structural path is released.

[0039] If the instance channel compression rate in the output of the optimized model is less than the set efficiency threshold, the response time of the instance in the real-time feedback data is lower than the preset response time threshold, and the service consistency result remains unchanged, then the scheduling priority of the instance will be increased, and the scheduling position of the service instance on the original execution node will remain unchanged.

[0040] If the response time in the real-time feedback data exceeds a set threshold for several consecutive time periods, the collaborative feature vector... If no mutation characteristics are detected, the service path is reconstructed, specifically by allocating additional channel resources and prioritizing the scheduling of hot standby nodes. Hot standby nodes refer to standby nodes that have been activated before service scheduling and are in state synchronization with the target service instance.

[0041] A cloud terminal intelligent service optimization system based on big data analysis according to an embodiment of the present invention includes the following modules:

[0042] The data acquisition and preprocessing module is used to collect data from cloud terminals and perform unified preprocessing.

[0043] The collaborative feature extraction module is used to input the structured input dataset into the temporal double-rotation nested network model and generate collaborative feature vectors;

[0044] The structure optimization execution module is used to perform generalized Hessian graph projection optimization based on collaborative feature vectors, and adjust the nesting depth and channel configuration.

[0045] The strategy generation module is used to generate cloud terminal resource scheduling strategies based on the structurally optimized model, including service instance mapping relationships and scheduling priorities.

[0046] The scheduling and execution module is used to deploy resource scheduling policies to the cloud service control module and control the task scheduling and load migration of service instances.

[0047] The feedback update module is used to collect response time, resource utilization and service consistency results in real time during the execution process, and update and optimize the temporal double-rotation nested network model.

[0048] The beneficial effects of this invention are:

[0049] (1) The present invention adopts a time-domain double-rotation nested network model, using the main rotation channel to capture the periodic pattern of service requests and the inverse rotation channel to identify non-periodic mutation behavior. By fusing multi-scale residual structures to generate collaborative feature vectors, the model's ability to express complex behavior sequences of cloud terminals is effectively improved.

[0050] (2) This invention introduces a generalized Hessian mapping projection optimization algorithm to reconstruct the nesting depth and channel configuration in the model, so that the model structure can be adaptively adjusted according to the importance of the input features. This breaks through the limitations of existing methods with fixed structure and lack of feedback adjustment, and improves the accuracy and flexibility of resource modeling.

[0051] (3) Based on the output of the optimized model, the present invention dynamically generates service instance mapping relationship and scheduling priority parameters, and performs closed-loop update in combination with real-time feedback data, which solves the problem of static configuration of existing strategies and lack of adaptability to behavior changes, and effectively enhances the stability and service continuity of terminal task scheduling. Attached Figure Description

[0052] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0053] Figure 1 This is an overall flowchart of a cloud terminal intelligent service optimization method based on big data analysis proposed in this invention;

[0054] Figure 2 This is a schematic diagram of the temporal double-rotation nested network model of a cloud terminal intelligent service optimization method based on big data analysis proposed in this invention. Detailed Implementation

[0055] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0056] refer to Figure 1 and Figure 2 A cloud terminal intelligent service optimization method based on big data analysis includes the following steps:

[0057] S1. Collect data from cloud terminals, perform unified preprocessing, and generate a structured input dataset;

[0058] S2. Input the structured input dataset into the established temporal double-rotation nested network model. The temporal double-rotation nested network model includes the main rotation channel, the inverse rotation channel, and nested residual blocks, and generates collaborative feature vectors.

[0059] S3. Construct a graph structure, perform generalized Hessian graph projection optimization on the collaborative feature vectors, calculate the second-order gradient response, optimize the nesting depth and channel configuration inside the temporal double-rotation nested network model, and obtain the structurally optimized model;

[0060] S4. Generate a cloud terminal resource scheduling strategy based on the optimized model. The cloud terminal resource scheduling strategy includes service instance mapping relationship and scheduling priority.

[0061] S5. Deploy resource scheduling strategies to the cloud service control module to control cloud terminals to perform task allocation and load migration operations;

[0062] S6. Real-time collection of response time, resource utilization and service consistency results of cloud terminal devices during the execution of the resource scheduling strategy, updating the structured and optimized model to achieve adaptive adjustment of strategy and structure.

[0063] This invention employs a temporal double-rotation nested network model by inputting structured input data into the model and combining it with a generalized Hessian graph projection optimization method. This enables deep feature extraction and adaptive structural modeling of complex behavioral data from cloud terminals, dynamically generating resource scheduling strategies and achieving model self-updating through feedback. Compared to traditional static configuration-based cloud service optimization schemes, this invention offers advantages such as high feature recognition accuracy, strong structural optimization capabilities, and intelligent and controllable scheduling strategies, effectively improving the response efficiency and resource utilization efficiency of cloud terminal systems.

[0064] In this embodiment, the unified preprocessing in step S1 includes performing timestamp alignment, missing value filling, format standardization, and outlier removal operations on the data collected from the cloud terminal.

[0065] This invention introduces timestamp alignment, missing value imputation, format standardization, and outlier removal mechanisms in the data preprocessing stage, providing a unified and reliable data foundation for subsequent feature extraction. This processing method improves the quality of the data stream input and avoids offset or structural mismatch problems caused by data anomalies during model training. Compared with the coarse data cleaning and temporal distortion problems in existing solutions, this invention has higher accuracy and robustness in terms of data consistency and input standardization.

[0066] In this embodiment, S2 specifically includes:

[0067] S21. Represent the structured input dataset as a matrix. The input is fed into the main rotation channel and processed by the rotating convolution kernel. Extract periodic features to obtain feature sequences , where the symbol This indicates a rotational convolution operation;

[0068] S22. Transform the structured input dataset Simultaneously input to the inverse rotation channel, through the inverse rotation convolution kernel Extract mutation features and obtain feature sequences ;

[0069] S23, The feature sequence and The features are merged into a single input feature set and fed into a multi-layered nested residual block structure. Each residual block contains information branch channels and feature conservation gating structures. Cross-scale feature fusion and residual connection operations are performed, and the fused collaborative feature vector is output. .

[0070] This invention designs a main rotating channel and an inverse rotating channel structure in the feature extraction stage, used for periodic behavior capture and sudden behavior detection, respectively. Combined with a nested residual structure, it achieves cross-scale feature fusion, ultimately outputting a collaborative feature vector. This structure overcomes the limitations of single-channel models in dynamic behavior modeling, enabling the system to accurately perceive the temporal characteristics of terminal behavior evolution, thus improving the model's discriminative and generalization abilities.

[0071] In this embodiment, the cross-scale feature fusion and residual connection operation of S23 specifically includes:

[0072] S231, merge the feature sequences and The synchronous input is fed into a multi-layer nested residual block structure, which contains several scale channel groups. Different channel groups extract feature information of the corresponding scale according to their respective set convolution receptive field size, and generate several sets of scale feature tensors.

[0073] S232. Set up an independent information branch channel for each set of scale feature tensors, and introduce a gating mechanism in each information branch channel to dynamically control whether the output of the feature tensor is passed according to the response characteristics of the current feature tensor.

[0074] S233. Fuse all the scale feature tensors after gating and adjustment, establish residual connections with the input features received by each scale channel group, and construct a nested residual structure with cross-scale fusion path.

[0075] S234. After fusion, output a unified collaborative feature vector. The collaborative feature vector As a synergistic feature representing periodic and mutable behavior.

[0076] This invention introduces multi-scale channels and information gating mechanisms to fuse features at various scales and constructs a cross-scale residual connection structure, enabling the model to retain low-level detailed features while possessing high-level semantic expression capabilities. This approach effectively enhances the network's ability to express heterogeneous behavioral features and improves the model's adaptability and structural stability in multi-type service behavior recognition tasks.

[0077] In this embodiment, S3 specifically includes:

[0078] S31, Combine the feature vectors Input is sent to the structure graph module to construct a graph structure corresponding to the model structure. ,in, This represents the set of structural nodes in a neural network. Indicates the connection relationship between nodes;

[0079] S32, in the diagram structure Above, based on collaborative feature vectors The forward propagation path is guided by which the total loss function of the model in the service response and resource prediction tasks is calculated. The second-order gradient response of each structural node to the total loss function is extracted, and the corresponding Hessian matrix is ​​constructed. ;

[0080] S33, Based on the Hessian matrix Local sensitivity changes in the data allow for the selection of the set of structural nodes to be adjusted. The structural update mapping relationship is generated by combining structural constraints.

[0081] S34. Apply the structure update mapping relationship to the graph structure. By adjusting the nesting depth and channel configuration within the temporal double-rotation nested network model through graph projection, the model with optimized structure is output.

[0082] This invention proposes a method to construct a structure graph based on collaborative feature vectors and identify sensitive structural regions using Hessian matrices, then update the model structure through graph projection. Compared to traditional structure search or static pruning methods, this approach offers superior optimization accuracy and interpretability, enabling real-time optimization of the model structure to adapt to changing service scenarios and improving the model's adaptability and execution efficiency for different service modes.

[0083] In this embodiment, S34, which adjusts the nesting depth and channel configuration within the temporal double-rotation nested network model via graph projection, specifically includes:

[0084] S341, In the diagram structure The set of nodes associated with the nested residual structure is identified in the middle. Each node represents a nested residual block, and the node attributes include the corresponding nesting level index and the number of channels.

[0085] S342, Based on Hessian matrix The response strength of a node in a set of nodes The nodes in the algorithm are sorted by their responses, and a subset of nodes with sensitivity below a preset value is selected. Used for structural adjustment;

[0086] S343, Targeting a subset of nodes Based on the preset minimum depth threshold and channel lower limit, the number of nested layers is pruned, or the channel configuration parameters are compressed and adjusted to generate a structural adjustment map.

[0087] S344. Project the structural adjustment diagram back to the original structural diagram. This forms the updated graph structure. The graph structure The corresponding temporal double-rotation nested network model is the structurally optimized model after adjusting the nesting depth and channel configuration.

[0088] This invention achieves automated adjustment of model nesting depth and channel number at the graph structure level by performing hierarchical response analysis and structural transformation on nested residual nodes and channel configuration nodes in the graph. Compared to manually designing model structures, this invention can dynamically reconstruct the structure based on the importance of input features, improving structural compactness and model performance ratio, making it particularly suitable for resource-sensitive cloud service scenarios.

[0089] In this embodiment, S4 specifically includes:

[0090] Cloud terminal resource scheduling strategy based on collaborative feature vector The optimized model output and real-time collected response data generation include the following scheduling rules:

[0091] If the collaborative feature vector If the periodic characteristic value represented by the main rotating channel increases in several consecutive time segments, and the channel configuration corresponding to the service instance in the optimized model has reached the current node's resource utilization limit, then the service instance will be migrated to another node, where the channel utilization rate of the other node is lower than the average level.

[0092] If the collaborative feature vector If the non-periodic features represented by the inverse rotation channel suddenly increase within a certain period of time, and the nesting depth of the instance is compressed to the minimum value in the output of the optimized model, then the scheduling frequency of the service instance is reduced, and the occupied part of the structural path is released.

[0093] If the instance channel compression rate in the output of the optimized model is less than the set efficiency threshold, the response time of the instance in the real-time feedback data is lower than the preset response time threshold, and the service consistency result remains unchanged, then the scheduling priority of the instance will be increased, and the scheduling position of the service instance on the original execution node will remain unchanged.

[0094] If the response time in the real-time feedback data exceeds a set threshold for several consecutive time periods, the collaborative feature vector... If no mutation characteristics are detected, the service path is reconstructed, specifically by allocating additional channel resources and prioritizing the scheduling of hot standby nodes. Hot standby nodes refer to standby nodes that have been activated before service scheduling and are in state synchronization with the target service instance.

[0095] The resource scheduling strategy proposed in this invention can form dynamic scheduling rules based on collaborative feature vectors and structural optimization results, combined with real-time feedback data, supporting control strategies such as service instance migration, frequency adjustment, and priority updates. Compared with existing fixed rules or static allocation strategies, the strategy of this invention has stronger context awareness and execution flexibility, significantly improving system response efficiency and service continuity under multi-terminal collaboration.

[0096] A cloud terminal intelligent service optimization system based on big data analysis according to an embodiment of the present invention includes the following modules:

[0097] The data acquisition and preprocessing module is used to collect data from cloud terminals and perform unified preprocessing.

[0098] The collaborative feature extraction module is used to input the structured input dataset into the temporal double-rotation nested network model and generate collaborative feature vectors;

[0099] The structure optimization execution module is used to perform generalized Hessian graph projection optimization based on collaborative feature vectors, and adjust the nesting depth and channel configuration.

[0100] The strategy generation module is used to generate cloud terminal resource scheduling strategies based on the structurally optimized model, including service instance mapping relationships and scheduling priorities.

[0101] The scheduling and execution module is used to deploy resource scheduling policies to the cloud service control module and control the task scheduling and load migration of service instances.

[0102] The feedback update module is used to collect response time, resource utilization and service consistency results in real time during the execution process, and update and optimize the temporal double-rotation nested network model.

[0103] The cloud terminal intelligent service optimization system constructed in this invention consists of six functional modules. These modules are logically clear and have tight data interfaces, enabling closed-loop control of the entire process from data collection, feature extraction, and structural optimization to strategy generation and feedback updates. This system architecture possesses modular deployment capabilities and high scalability, adapting to various cloud terminal deployment environments and providing system-level support for achieving intelligent and adaptive cloud service optimization.

[0104] Example 1:

[0105] To verify the feasibility of this invention in practice, it was applied to a user request management and resource scheduling optimization scenario of a large-scale distributed cloud service platform. This platform provides core capabilities such as computing scheduling, service invocation, and interactive response to hundreds of thousands of active cloud terminals daily. Under the traditional service operation architecture, this platform adopts a fixed weight allocation strategy and a static resource mapping method. With the continuous growth of service load and the diversification of user behavior patterns, problems such as unstable response latency, uneven resource utilization, and the inability to quickly identify abnormal behavior have gradually emerged. Especially in scenarios with sudden traffic surges, multiple terminal nodes experience service drift due to congestion in scheduling paths or lag in the structural model response, leading to a decline in terminal experience and deterioration of system stability.

[0106] In the application of this invention, a data acquisition and preprocessing module was first deployed to continuously collect operation logs, user access behavior, and system communication data from 112 cloud terminal devices in a selected cluster. Through unified time alignment and anomaly removal, an average of approximately 97.3 MB of effective structured data was generated per device per day. Subsequently, the data was input into the temporal dual-rotation nested network model proposed in this invention. The main rotation channel extracts periodic behavior patterns, while the inverse rotation channel identifies abrupt behavioral features. After fusion, a collaborative feature vector is output, achieving a "behavioral mutation prediction accuracy" of 91.4% on the training set, significantly outperforming traditional LSTM and 1D-CNN models.

[0107] Building upon this, a generalized Hessian graph projection optimization algorithm was invoked to adjust the network structure driven by collaborative feature vectors. The nesting depth was dynamically adjusted from the original fixed 6 layers to switch between 396 channels, resulting in a 23.8% reduction in the average latency of the model during the inference phase. Simultaneously, the output of the structure-optimized model was used in the resource scheduling strategy generation module, which automatically constructs service instance mapping relationships and a node scheduling priority list, supporting dynamic transfer of execution nodes based on channel utilization.

[0108] After deployment, performance was compared with the original system under the same test window. In high-concurrency request load scenarios, the average system response time decreased from 428ms to 286ms, a reduction of 33.2%; resource utilization increased from 68.7% to 82.9%; scheduling failure rate decreased from 3.6% to 1.1%; and service consistency score remained stable above 97%. Furthermore, in terms of sudden behavior recognition, the system can complete partial reconstruction of the model structure and policy update within an average of 2.4 seconds after a sudden increase in behavior, far faster than the traditional periodic refresh mechanism (average 11.6 seconds), ensuring the continuity of the scheduling link and the real-time nature of behavior response.

[0109] To further demonstrate the system's performance, the following table summarizes the changes in key performance indicators before and after deployment over different time periods:

[0110] Table 1: Comparison of Cloud Terminal Optimized Scheduling Performance Indicators Based on the System of the Invention

[0111] Test section Average response time (ms) Resource utilization rate (%) Service consistency score (%) Scheduling failure rate (%) Average response time (s) for identifying sudden behavior Original system - morning peak 445 66.2 92.5 3.9 12.4 Original system - stable at midday 392 70.1 94.3 2.8 11.2 Original system - evening fluctuations 428 68.7 93.6 3.6 11.6 This invention - morning peak 298 80.4 97.1 1.3 2.7 This invention - midday stability 273 84.6 97.6 1.0 2.1 This invention - Evening fluctuations 286 82.9 97.3 1.1 2.4

[0112] Overall, the present invention demonstrates significantly superior performance compared to the original system under various load conditions. Firstly, regarding average response time, the system significantly shortens the service response cycle across all three time periods, particularly during the morning peak hours, reducing it from 445 milliseconds to 298 milliseconds, a decrease of 33.0%. This effectively improves the immediacy of user request processing and reduces service lag and queuing.

[0113] Regarding resource utilization, the system of this invention achieves dynamic and efficient allocation of cloud terminal node resources through structural adaptation and scheduling strategy optimization. Data in the table shows that the resource utilization rate of the system of this invention increased to 84.6% during the stable midday period, an improvement of more than 14 percentage points compared to the original system's 70.1%, demonstrating that the system also possesses excellent resource integration capabilities under low-pressure conditions.

[0114] Service consistency score is a core indicator of system stability and task accuracy. The original system dropped to 92.5% during peak periods, while the present invention still maintained above 97.1%, indicating that the model of the present invention has stronger fault tolerance to sudden behavior and load fluctuations, ensuring the continuity of service logic and output quality.

[0115] Regarding scheduling failure rate, the system of this invention significantly reduces the error and interruption rate during the scheduling process through dynamic mapping and channel compression mechanisms. The original system had a failure rate as high as 3.6% during the evening fluctuation period, while the failure rate of this invention was only 1.1% during the same period, demonstrating its higher ability to stably deliver tasks and its suitability for complex and ever-changing service request environments.

[0116] More notably, this invention demonstrates extremely high real-time performance in responding to sudden events. As shown in the table, when a sudden event occurs, the system of this invention completes model structure reconstruction and policy response in an average of only 2.4 seconds, compared to the original system's 11.6-second delay. This effectively shortens the response link between identification and scheduling, greatly improving the system's ability to perceive and control sudden high-frequency events.

[0117] In summary, through verification by multi-dimensional indicators, this invention can achieve comprehensive advantages such as more accurate behavior perception, more intelligent model adjustment, more dynamic scheduling strategy, and more stable execution effect in complex cloud terminal environments, and has strong engineering practicality and broad promotion value.

[0118] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A cloud terminal intelligent service optimization method based on big data analysis, characterized in that, Includes the following steps: S1. Collect data from cloud terminals, perform unified preprocessing, and generate a structured input dataset; S2. Input the structured input dataset into the established temporal double-rotation nested network model. The temporal double-rotation nested network model includes the main rotation channel, the inverse rotation channel, and nested residual blocks, and generates collaborative feature vectors. Specifically, it includes: S21. Represent the structured input dataset as a matrix. The input is fed into the main rotation channel and processed by the rotating convolution kernel. Extract periodic features to obtain feature sequences , where the symbol This indicates a rotational convolution operation; S22. Transform the structured input dataset Simultaneously input to the inverse rotation channel, through the inverse rotation convolution kernel Extract mutation features and obtain feature sequences ; S23, The feature sequence and The features are merged into a single input feature set and fed into a multi-layered nested residual block structure. Each residual block contains information branch channels and feature conservation gating structures. Cross-scale feature fusion and residual connection operations are performed, and the fused collaborative feature vector is output. ; S3. Construct the graph structure, perform generalized Hessian graph projection optimization on the collaborative feature vectors, calculate the second-order gradient response, optimize the nesting depth and channel configuration within the temporal double-rotation nested network model, and obtain the structurally optimized model; specifically including: S31, Combine the feature vectors Input is sent to the structure graph module to construct a graph structure corresponding to the model structure. ,in, This represents the set of structural nodes in a neural network. Indicates the connection relationship between nodes; S32, in the diagram structure Above, based on collaborative feature vectors The forward propagation path is guided to calculate the total loss function of the model in the service response and resource prediction tasks, extract the second-order gradient response of each structural node to the total loss function, and construct the corresponding Hessian matrix. S33, Based on the Hessian matrix Local sensitivity changes in the data allow for the selection of the set of structural nodes to be adjusted. The structural update mapping relationship is generated by combining structural constraints. S34. Apply the structure update mapping relationship to the graph structure. By adjusting the nesting depth and channel configuration within the temporal double-rotation nested network model through graph projection, the model with optimized structure is output. S4. Generate a cloud terminal resource scheduling strategy based on the optimized model. The cloud terminal resource scheduling strategy includes service instance mapping relationship and scheduling priority. S5. Deploy resource scheduling strategies to the cloud service control module to control cloud terminals to perform task allocation and load migration operations; S6. Real-time collection of response time, resource utilization and service consistency results of cloud terminal devices during the execution of the resource scheduling strategy, updating the structured and optimized model to achieve adaptive adjustment of strategy and structure.

2. The cloud terminal intelligent service optimization method based on big data analysis according to claim 1, characterized in that, The unified preprocessing in step S1 includes performing timestamp alignment, missing value imputation, format standardization, and outlier removal operations on the data collected from the cloud terminal.

3. The cloud terminal intelligent service optimization method based on big data analysis according to claim 2, characterized in that, The cross-scale feature fusion and residual connection operation in S23 specifically includes: S231, merge the feature sequences and The synchronous input is fed into a multi-layer nested residual block structure, which contains several scale channel groups. Different channel groups extract feature information of the corresponding scale according to their respective set convolution receptive field size, and generate several sets of scale feature tensors. S232. Set up an independent information branch channel for each set of scale feature tensors, and introduce a gating mechanism in each information branch channel to dynamically control whether the output of the feature tensor is passed according to the response characteristics of the current feature tensor. S233. Fuse all the scale feature tensors after gating and adjustment, establish residual connections with the input features received by each scale channel group, and construct a nested residual structure with cross-scale fusion path. S234. After fusion, output a unified collaborative feature vector. The collaborative feature vector As a synergistic feature representing periodic and mutable behavior.

4. The cloud terminal intelligent service optimization method based on big data analysis according to claim 3, characterized in that, The S34 method, which adjusts the nesting depth and channel configuration within the temporal double-rotation nested network model via graph projection, specifically includes: S341, In the diagram structure The set of nodes associated with the nested residual structure is identified in the middle. Each node represents a nested residual block, and the node attributes include the corresponding nesting level index and the number of channels. S342, Based on Hessian matrix The response strength of a node in a set of nodes The nodes in the algorithm are sorted by their responses, and a subset of nodes with sensitivity below a preset value is selected. Used for structural adjustment; S343, Targeting a subset of nodes Based on the preset minimum depth threshold and channel lower limit, the number of nested layers is pruned, or the channel configuration parameters are compressed and adjusted to generate a structural adjustment map. S344. Project the structural adjustment diagram back to the original structural diagram. This forms the updated graph structure. The graph structure The corresponding temporal double-rotation nested network model is the structurally optimized model after adjusting the nesting depth and channel configuration.

5. The cloud terminal intelligent service optimization method based on big data analysis according to claim 4, characterized in that, S4 specifically includes: Cloud terminal resource scheduling strategy based on collaborative feature vector The optimized model output and real-time collected response data generation include the following scheduling rules: If the collaborative feature vector If the periodic characteristic value represented by the main rotating channel increases in several consecutive time segments, and the channel configuration corresponding to the service instance in the optimized model has reached the current node's resource utilization limit, then the service instance will be migrated to another node, where the channel utilization rate of the other node is lower than the average level. If the collaborative feature vector If the non-periodic features represented by the inverse rotation channel suddenly increase within a certain period of time, and the nesting depth of the instance is compressed to the minimum value in the output of the optimized model, then the scheduling frequency of the service instance is reduced, and the occupied part of the structural path is released. If the instance channel compression rate in the output of the optimized model is less than the set efficiency threshold, the response time of the instance in the real-time feedback data is lower than the preset response time threshold, and the service consistency result remains unchanged, then the scheduling priority of the instance will be increased, and the scheduling position of the service instance on the original execution node will remain unchanged. If the response time in the real-time feedback data exceeds a set threshold for several consecutive time periods, the collaborative feature vector... If no mutation characteristics are detected, the service path is reconstructed, specifically by allocating additional channel resources and prioritizing the scheduling of hot standby nodes. Hot standby nodes refer to standby nodes that have been activated before service scheduling and are in state synchronization with the target service instance.

6. A cloud terminal intelligent service optimization system based on big data analysis, applied to the cloud terminal intelligent service optimization method based on big data analysis as described in any one of claims 1 to 5, characterized in that, Includes the following modules: The data acquisition and preprocessing module is used to collect data from cloud terminals and perform unified preprocessing. The collaborative feature extraction module is used to input the structured input dataset into the temporal double-rotation nested network model and generate collaborative feature vectors; The structure optimization execution module is used to perform generalized Hessian graph projection optimization based on collaborative feature vectors, and adjust the nesting depth and channel configuration. The strategy generation module is used to generate cloud terminal resource scheduling strategies based on the structurally optimized model, including service instance mapping relationships and scheduling priorities. The scheduling and execution module is used to deploy resource scheduling policies to the cloud service control module and control the task scheduling and load migration of service instances. The feedback update module is used to collect response time, resource utilization and service consistency results in real time during the execution process, and update and optimize the temporal double-rotation nested network model.

Images