Intelligent computing power integration service management method and platform based on cloud side-end cooperation

Abstract

The invention discloses an intelligent computing power integration service management method and platform based on cloud side-end cooperation, and relates to the technical field of computing power integration management, and the method comprises the steps: sensing the computing power resource state of a side-end cooperation port and terminal equipment in real time at a cloud controller, and building a resource topological graph; a dynamic task demand is introduced, and a computing power scheduling vector is generated; carrying out lightweight segmentation on the cloud training model, and determining adjacent edge nodes under hierarchical limitation to form a regional elastic cluster; and deploying a digital twin simulation engine rehearsal computing power distribution scene, establishing a heterogeneous resource pooling mechanism, and carrying out computing power integration service management. The technical problems of low management efficiency and insufficient utilization rate of heterogeneous computing power resources in the prior art are solved, and the technical effects of realizing efficient management of intelligent computing power integration services and improving the utilization rate of the heterogeneous computing power resources are achieved.

Description

Technical Field

[0001] The present invention relates to the field of computing power integration management technology, and specifically to an intelligent computing power integration service management method and platform based on cloud-edge-end collaboration. Background Art

[0002] With the rapid development of cloud computing, edge computing, and terminal devices, computing power demands are becoming increasingly diverse and dynamic. Integrated management of heterogeneous computing resources (such as GPUs, CPUs, and FPGAs) is crucial for improving service efficiency. However, existing technologies suffer from issues such as delayed computing resource awareness, inflexible scheduling strategies, insufficient load balancing, and poor energy consumption control. These issues lead to low utilization of heterogeneous computing resources and high task processing latency, making it difficult to meet the dynamic task demands in complex scenarios.

[0003] The existing technology has the technical problems of low efficiency in heterogeneous computing resource management and insufficient utilization. Summary of the Invention

[0004] This application provides an intelligent computing power integrated service management method and platform based on cloud-edge-end collaboration, which is used to solve the technical problems of low efficiency and insufficient utilization of heterogeneous computing power resource management in the existing technology.

[0005] In view of the above problems, this application provides an intelligent computing power integrated service management method and platform based on cloud-edge-end collaboration.

[0006] The first aspect of the present application provides a method for managing intelligent computing power integration services based on cloud-edge-end collaboration, the method comprising:

[0007] In the cloud controller, the computing power resource status of the edge collaborative port and the terminal device is perceived in real time to build a resource topology map; based on the resource topology map, dynamic task requirements are introduced to generate a computing power scheduling vector that includes computing power load balancing weights, delay sensitivity indicators and energy consumption constraints; according to the computing power scheduling vector, the cloud training model corresponding to the cloud controller is lightweight segmented, the edge node reasoning capabilities are dynamically matched, and the adjacent edge nodes under the hierarchical limitations are determined to form a regional elastic cluster; a digital twin simulation engine is deployed to pre-simulate the computing power allocation scenario, combined with reinforcement learning to dynamically correct the resource fragmentation rate, establish a heterogeneous resource pooling mechanism, and dynamically divide the heterogeneous computing power resources within the regional elastic cluster on demand to perform computing power integration service management.

[0008] The second aspect of this application provides an intelligent computing power integration service management platform based on cloud-edge-end collaboration, the platform comprising:

[0009] A resource topology map construction module is used to perceive the computing power resource status of edge collaborative ports and terminal devices in real time in the cloud controller and build a resource topology map; a dynamic task demand introduction module is used to introduce dynamic task demands based on the resource topology map, and generate a computing power scheduling vector containing computing power load balancing weights, delay sensitivity indicators and energy consumption constraints; a regional elastic cluster composition module is used to perform lightweight segmentation of the cloud training model corresponding to the cloud controller according to the computing power scheduling vector, dynamically match the edge node reasoning capabilities, and determine the adjacent edge nodes under the hierarchical limitations to form a regional elastic cluster; a computing power integration service management module is used to deploy a digital twin simulation engine to pre-simulate computing power allocation scenarios, combine reinforcement learning to dynamically correct resource fragmentation rates, establish a heterogeneous resource pooling mechanism, and dynamically divide the heterogeneous computing power resources within the regional elastic cluster on demand to perform computing power integration service management.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] The cloud controller perceives the computing resource status of edge collaborative ports and terminal devices in real time, constructs a resource topology map, introduces dynamic task requirements, and generates computing scheduling vectors. It then performs lightweight segmentation on the cloud training model corresponding to the cloud controller, determining the adjacent edge nodes within the hierarchical constraints to form a regional elastic cluster. A digital twin simulation engine is deployed to rehearse computing power allocation scenarios, establish a heterogeneous resource pooling mechanism, and dynamically divide the heterogeneous computing power resources within the regional elastic cluster on demand for computing power integration service management. This achieves the technical effect of achieving efficient management of intelligent computing power integration services and improving the utilization of heterogeneous computing power resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 A flowchart of a method for managing intelligent computing power integration services based on cloud-edge-end collaboration provided in an embodiment of the present application;

[0014] Figure 2 A structural diagram of the intelligent computing power integrated service management platform based on cloud-edge-end collaboration provided in an embodiment of the present application.

[0015] Explanation of the accompanying drawings: resource topology map building module 10, dynamic task demand introduction module 20, regional elastic cluster composition module 30, computing power integration service management module 40. DETAILED DESCRIPTION

[0016] This application provides an intelligent computing power integrated service management method and platform based on cloud-edge-end collaboration to solve the technical problems of low efficiency and insufficient utilization of heterogeneous computing power resource management in the existing technology.

[0017] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative work are within the scope of protection of this application.

[0018] Example 1, as Figure 1 As shown, this application provides an intelligent computing power integrated service management method based on cloud-edge-end collaboration, the method comprising:

[0019] Step S100: In the cloud controller, the computing resource status of the edge collaborative port and the terminal device is perceived in real time, and a resource topology map is built.

[0020] Specifically, the cloud controller monitors the computing resource status of edge collaborative ports and terminal devices in real time. This status information covers device load, network bandwidth, energy consumption data, etc., and at the same time associates relevant data in the user behavior portrait library and device topology data, and integrates the perceived resource information according to device connection relationships, hierarchical structures and resource attributes. It constructs a resource topology map that can clearly present the resource distribution, association relationships and potential interaction paths between edge devices and collaborative ports, providing comprehensive and real-time resource basic data support for subsequent steps such as introducing dynamic task requirements and generating computing power scheduling vectors.

[0021] Step S200: Based on the resource topology map, dynamic task requirements are introduced to generate a computing power scheduling vector including computing power load balancing weights, delay sensitivity indicators and energy consumption constraints.

[0022] Specifically, the load forecast weight is first determined by connecting to the user behavior portrait library, and the partition forecast benchmark curve is formulated in combination with the equipment topology data, and dynamic task requirements are introduced on this basis; in the process of generating the computing power scheduling vector, the user's electricity consumption behavior time series data and the equipment infrared thermal imaging data are integrated, and cross-dimensional features are extracted through the spatiotemporal attention mechanism. The weights of the multi-region load forecast model are aggregated based on the federated learning framework, and the harmonic distortion rate compensation parameters are dynamically optimized in combination with the cross-dimensional features. At the same time, a dynamic reactive power compensation strategy is set and combined with LSTM to predict the grid voltage fluctuation trend. Finally, a computing power scheduling vector is formed that includes computing power load balancing weights, delay sensitivity indicators and energy consumption constraints. The deployed hardware-accelerated edge nodes will process the relevant parameters in parallel to ensure timely response.

[0023] Step S300: Based on the computing power scheduling vector, the cloud training model corresponding to the cloud controller is lightweight segmented, the edge node reasoning capability is dynamically matched, and the adjacent edge nodes under the hierarchical limitation are determined to form a regional elastic cluster.

[0024] Specifically, according to the generated computing power scheduling vector, the user's electricity consumption behavior time series data and equipment infrared thermal imaging data are first integrated, and cross-dimensional features are extracted through the spatiotemporal attention mechanism. Then, based on the federated learning framework, the weights of the multi-region load forecasting model are aggregated. The harmonic distortion rate compensation parameters are dynamically optimized in combination with the cross-dimensional features, and the cloud model parameters are compressed through the lightweight knowledge distillation architecture to complete the lightweight segmentation of the cloud training model corresponding to the cloud controller; at the same time, a dynamic reactive power compensation strategy is set to generate real-time compensation instructions based on inductive load characteristics and capacitive load characteristics, and the LSTM is combined to predict the grid voltage fluctuation trend, and hardware-accelerated edge nodes are deployed to parallelly process harmonic suppression parameters to ensure that the dynamic compensation response time meets the time limit, thereby dynamically matching the edge node reasoning capabilities, and then determining the adjacent edge nodes under the hierarchical limitation to form a regional elastic cluster.

[0025] Step S400: Deploy a digital twin simulation engine to rehearse computing power allocation scenarios, combine reinforcement learning to dynamically correct resource fragmentation rates, establish a heterogeneous resource pooling mechanism, dynamically divide heterogeneous computing power resources within the regional elastic cluster on demand, and perform computing power integration service management.

[0026] Specifically, a digital twin simulation engine is deployed to rehearse computing power allocation scenarios. During this process, network bandwidth thresholds, latency tolerance limits, and energy consumption increment limits are inserted along the load transfer path. A multi-objective optimization solution is then implemented with load balancing, transfer costs, and service quality as goals. Simulating sensor noise interference patterns using a generative adversarial network dynamically optimizes the load feature extraction threshold. Historical equipment failure data is then correlated with this threshold to correct the confidence interval of the partition prediction baseline curve using Monte Carlo simulation. Equipment operating data is then compared with preset risk thresholds. If exceeded, an alert is triggered and feedback is sent to the cloud controller. Furthermore, reinforcement learning is used to dynamically adjust the resource fragmentation rate to establish a heterogeneous resource pooling mechanism. The allocation of tasks to edge nodes is dynamically adjusted based on the device energy efficiency ratio, combined with photovoltaic output and task processing progress. The energy consumption of each node after use is monitored. The system supports on-demand dynamic partitioning of GPU, CPU, and FPGA computing power through the integration of hardware accelerators and containerization. Energy efficiency of each type of computing power is quantitatively evaluated based on energy consumption data, and its allocation is optimized within the regional elastic cluster. This allows for on-demand dynamic partitioning of heterogeneous computing power resources within the regional elastic cluster, enabling integrated computing power service management.

[0027] In one possible implementation, step S300 further includes:

[0028] Step S310: Integrate user electricity usage behavior time series data and device infrared thermal imaging data, and extract cross-dimensional features through the spatiotemporal attention mechanism.

[0029] Step S320: Aggregate the weights of the multi-region load forecasting model based on the federated learning framework, dynamically optimize the harmonic distortion rate compensation parameters in combination with the cross-dimensional features, and compress the cloud-based model parameters through a lightweight knowledge distillation architecture.

[0030] Specifically, we first integrate the time series data of electricity consumption behavior generated by users at different time nodes (such as power load fluctuations, changes in power consumption duration, and other data recorded in time series) and the temperature distribution, heat dissipation efficiency and other status data captured by the equipment through infrared thermal imaging technology. Then, we use the spatiotemporal attention mechanism to deeply process these data. This mechanism can focus on the dynamic changes in electricity consumption behavior in the time dimension, and pay attention to the physical characteristics presented by the thermal imaging of the equipment in the spatial dimension. Through cross-dimensional correlation analysis and feature mining, we extract key cross-dimensional features that can comprehensively reflect the user's electricity consumption habits and the operating status of the equipment, providing basic data support for the lightweight processing and parameter optimization of the subsequent cloud-based training model.

[0031] Based on the federated learning framework, the FedAvg algorithm is used to aggregate the weights of multi-regional load forecasting models. After each regional node trains the model locally using the gradient descent algorithm, it only uploads the model parameters to the cloud. The cloud then performs weighted averaging of the parameters through FedAvg to obtain a global model. When dynamically optimizing the harmonic distortion rate compensation parameters based on cross-dimensional features, the random forest algorithm is introduced to explore the mapping relationship between features and compensation parameters. The parameters are adjusted through multiple rounds of iterations to minimize the harmonic distortion rate. In the process of compressing the cloud model parameters, the knowledge distillation algorithm is used, with the complex cloud model as the teacher model and the lightweight model on the edge as the student model. The output distributions of the two are aligned using the KL divergence loss function. At the same time, the L1 regularization algorithm is used to sparsify the student model parameters to achieve effective compression of the model parameters.

[0032] In one possible implementation, step S300 further includes:

[0033] Step S330: Set a dynamic reactive power compensation strategy, generate real-time compensation instructions based on inductive load characteristics and capacitive load characteristics, and combine LSTM to predict grid voltage fluctuation trends.

[0034] Step S340: At the same time, hardware-accelerated edge nodes are deployed to process harmonic suppression parameters in parallel so that the dynamic compensation response time meets the response time limit.

[0035] Specifically, by setting up a dynamic reactive power compensation strategy, the system collects and analyzes the inductive load characteristics (such as inductance parameters and power factor lag) and capacitive load characteristics (such as capacitance parameters and power factor lead) in the circuit in real time. Based on these characteristic data, corresponding real-time compensation instructions are generated to achieve dynamic balance of the grid's reactive power. At the same time, the LSTM (Long Short-Term Memory) model is introduced, leveraging its advantages in processing time series data. Combined with historical grid voltage data training models, the model can then predict the fluctuation trend of the grid voltage, enabling the reactive power compensation strategy to respond to voltage changes in advance, improving the accuracy and foresight of compensation.

[0036] While implementing the dynamic reactive power compensation strategy, edge nodes equipped with hardware acceleration components such as FPGAs are deployed. Leveraging hardware-level parallel processing capabilities, these nodes can efficiently parallelize and adjust harmonic suppression parameters (such as harmonic frequency, amplitude, and phase compensation) in real time. Hardware acceleration technology shortens the parameter processing cycle, ensuring that the dynamic compensation response time, from receiving compensation requests to generating specific suppression parameters, is within preset time limits. This ensures the timely implementation of harmonic suppression actions, synergizes with the dynamic reactive power compensation strategy, and improves the overall efficiency of grid harmonic management.

[0037] In one possible implementation, step S400 further includes:

[0038] Step S410: Connect to the user behavior profile library to determine the load prediction weight, and formulate a partition prediction benchmark curve based on the device topology data.

[0039] Step S420: Based on the partition prediction benchmark curve, a blockchain evidence storage unit is integrated, and the blockchain evidence storage unit is used to record the prediction model parameter update log.

[0040] Specifically, the pre-built user behavior portrait library is connected through an interface. The library stores behavioral data such as users' historical electricity usage time, power peak period, equipment start-up and shutdown patterns, etc. The weighted regression algorithm is used to evaluate the feature importance of these data to determine the contribution of different user behavior characteristics in load forecasting, that is, the load forecast weight; at the same time, the device topology data (including the physical connection relationship of the equipment, hierarchical distribution and regional affiliation information) is called, and the K-means clustering algorithm is used to partition the equipment and coverage area according to the similarity of power consumption characteristics. Combined with the historical load mean, fluctuation range and trend characteristics of each partition, the load change baseline of each partition is generated by fitting through the time series smoothing method, that is, the partition prediction baseline curve.

[0041] Based on the partition prediction benchmark curve, a blockchain evidence storage unit is integrated through the deployment of a consortium chain architecture. This unit uses a chain data structure and hash encryption algorithm to generate blocks for key information about prediction model parameter updates (including update timestamps, specific values ​​before and after parameter changes, and operation node identifiers), which are linked and stored in chronological order. At the same time, smart contracts are used to preset trigger conditions and recording rules for parameter updates. This ensures that after each model parameter adjustment, relevant log information is automatically uploaded to the chain for evidence storage, making the parameter update process tamper-proof and fully traceable, providing reliable data recording support for model parameter management for partition load forecasting.

[0042] In one possible implementation, step S400 further includes:

[0043] Step S430: Based on the computing power allocation scenario, insert the network bandwidth threshold, latency tolerance upper limit, and energy consumption increment limit under the load transfer path.

[0044] Step S440: Perform multi-objective optimization solution with load balancing, transfer cost, and service quality as the goals.

[0045] Specifically, based on the computing power allocation scenario previewed by the digital twin simulation engine, three key limiting parameters are precisely inserted for the load transfer path: the network bandwidth threshold, which is the maximum bandwidth value allowed for data transmission along this path to ensure that data transmission is not congested; the latency tolerance upper limit, which specifies the maximum acceptable delay time during the load transfer process to ensure the timeliness of service response; and the energy consumption increment limit, which sets the maximum energy consumption increase allowed for load transfer compared to the original state, taking into account energy conservation needs. The insertion of these parameters establishes clear constraint boundaries for the computing power allocation scenario and provides the basic conditions for subsequent multi-objective optimization solutions.

[0046] When solving multi-objective optimization problems, NSGA-II (Non-Dominated Sorting Genetic Algorithm II) is employed, with load balancing, transfer costs, and service quality as optimization objectives. A fitness function is constructed in conjunction with constraints such as network bandwidth thresholds, latency tolerance limits, and energy consumption increment limits. The algorithm generates multiple possible computing power allocation schemes by initializing the population. For each scheme, specific values ​​for the three objectives are calculated. Non-dominated sorting is used to divide the schemes into different levels, and population diversity is maintained using congestion distance. Through iterative optimization using genetic operations such as selection, crossover, and mutation, a Pareto-optimal solution set is ultimately obtained. Each solution corresponds to a computing power allocation scheme that achieves a balance between the three objectives, ensuring the coordinated optimization of load balancing, cost control, and service quality while satisfying the constraints.

[0047] In one possible implementation, step S400 further includes:

[0048] Step S450: Dynamically optimize the load feature extraction threshold based on the generative adversarial network to simulate the sensor noise interference pattern.

[0049] Step S460: Use the digital twin simulation engine to correlate equipment failure history data with load feature extraction thresholds for risk assessment.

[0050] Specifically, a noise simulation model is constructed based on a generative adversarial network. The generator simulates various noise interference patterns that may occur in sensors under different operating conditions. The discriminator is used to identify the authenticity of the generated noise. The accuracy of the noise simulation is improved through multiple rounds of adversarial training. The simulated noise interference pattern is combined with the actual collected load data to simulate the load data input scenario in a complex noise environment. On this basis, the threshold of load feature extraction is dynamically adjusted to ensure that key load characteristics can still be accurately captured in the presence of noise interference, thereby optimizing the effectiveness of load feature extraction and anti-interference ability.

[0051] Build a digital twin simulation engine, construct a virtual mapping model consistent with the physical equipment and system, import the equipment historical fault database (including data such as fault type, occurrence time, related operating parameters and impact degree) and the load feature extraction threshold optimized by the generative adversarial network; use the engine to simulate the operating status of the equipment under different threshold settings, establish a correlation model between the probability of failure and the feature extraction threshold, quantitatively analyze the impact of threshold changes on failure risks, and output a risk assessment report to provide data support for subsequent risk warnings and threshold adjustments.

[0052] In one possible implementation, step S460 further includes:

[0053] Step S461: Based on Monte Carlo simulation, dynamically correct the confidence interval of the partition prediction reference curve.

[0054] Step S462: The revised confidence interval is combined with the equipment operation data to compare with the preset risk threshold. If it exceeds the threshold, the early warning mechanism is triggered, and the risk assessment result is fed back to the cloud controller.

[0055] Specifically, using the Monte Carlo simulation method, multiple sets of simulated load data are generated by setting random variables related to load (such as fluctuations in user power consumption and changes in equipment operating status). Based on the probability distribution of these variables, repeated sampling is performed to generate multiple sets of simulated load data. This simulated data is then compared and analyzed with the original partition prediction baseline curve. The deviation between the curve's predicted and simulated values ​​is calculated, and the upper and lower limits of the baseline curve's confidence interval are dynamically adjusted. This revised confidence interval more accurately reflects the range of possible actual load fluctuations, improving the interval's reliability and adaptability.

[0056] Using the confidence interval of the Monte Carlo simulation-corrected partition prediction baseline curve as a reference range, the system collects real-time equipment operating data (such as real-time load values, equipment temperature, and energy consumption indicators) and compares this data against preset risk thresholds. If the equipment operating data exceeds the boundaries of the corrected confidence interval and the preset risk threshold, the early warning mechanism is immediately triggered, issuing warnings through audio and visual alarms, information push, and other means. At the same time, the detailed results of this risk assessment (including exceeding threshold data, risk level, and possible impact range) are fed back to the cloud controller in real time, allowing the cloud to further adjust computing power scheduling or strategies based on this feedback information.

[0057] In one possible implementation, step S400 further includes:

[0058] Step S470: Based on the equipment energy efficiency ratio, combined with photovoltaic output and task processing progress, dynamically adjust the distribution ratio of tasks on each edge node.

[0059] Step S480: At the same time, monitor the energy consumption of each edge node after computing power is used, and update the heterogeneous resource pooling mechanism.

[0060] Specifically, a weighted greedy algorithm is used. Weight coefficients are first set for the device's energy efficiency ratio, photovoltaic output, and task processing progress (e.g., energy efficiency weight 0.5, photovoltaic output weight 0.3, and task processing progress weight 0.2). The three indicators of each edge node are then standardized (mapped to the range of 0-1). A weighted formula (node ​​score = energy efficiency ratio × 0.5 + photovoltaic output × 0.3 + task processing progress adaptability × 0.2) is used to calculate the overall score of each node. Tasks are dynamically allocated based on the score proportion: the task allocation proportion of a node = the node's score / the total score of all nodes. This ensures that tasks are allocated to nodes with high energy efficiency, stable photovoltaic output, and adaptability to task processing progress.

[0061] Energy consumption monitoring sensors and data acquisition modules are deployed at each edge node to record data such as power consumption and energy consumption fluctuations of the node in the process of processing computing tasks in real time, and transmit the data to the cloud controller through the edge collaborative port; the cloud controller analyzes the received energy consumption data and calculates energy efficiency indicators such as unit computing power energy consumption based on the node's computing power type (such as GPU, CPU, FPGA) and task processing volume; based on these indicators, the dynamic segmentation of various types of computing power is supported through hardware acceleration cards and containerization fusion technology, and the ratio of different types of computing power in the regional elastic cluster is adjusted based on the results of energy efficiency quantitative evaluation, thereby updating the heterogeneous resource pooling mechanism to ensure that resource allocation is more in line with energy efficiency requirements.

[0062] In one possible implementation, step S480 further includes:

[0063] Step S481: Through the integration of hardware acceleration cards and containerization, support for dynamic on-demand division of GPU computing power, CPU computing power, and FPGA computing power.

[0064] Step S482: At the same time, the energy efficiency of GPU computing power, CPU computing power, and FPGA computing power is quantitatively evaluated based on the energy consumption data, and the ratio is optimized in the regional elastic cluster.

[0065] Specifically, hardware accelerator cards are integrated with containerization technology to build a unified computing power scheduling framework. Hardware accelerator cards provide efficient hardware support for different types of computing power, such as GPUs, CPUs, and FPGAs, while containerization technology implements virtualization, encapsulation, and isolation of these computing power resources, allowing various types of computing power to be flexibly divided into multiple independent computing units. Through this fusion approach, the system can dynamically and accurately divide GPU computing power, CPU computing power, and FPGA computing power in real time based on the actual task's demand for different types of computing power, meeting the flexible needs of diverse tasks in terms of computing power type and scale, and improving the utilization efficiency of heterogeneous computing power resources.

[0066] Build an energy efficiency quantitative evaluation model to collect energy consumption data (such as power consumption, energy consumption fluctuation, etc.) of GPU, CPU, and FPGA in real time during task processing, and calculate quantitative indicators such as unit energy consumption computing power value based on their computing power output data; score and rank the energy efficiency of the three types of computing power based on these indicators, and then combine the task types and computing power requirements within the regional elastic cluster to dynamically adjust the proportion of the three in the cluster through an intelligent scheduling algorithm, giving priority to allocating high-energy-consuming tasks to computing power types with high energy efficiency scores, while reducing the load proportion of computing power with lower energy efficiency, so as to optimize the ratio of heterogeneous computing power within the cluster and improve overall energy efficiency.

[0067] Example 2 is based on the same inventive concept as the intelligent computing power integrated service management method based on cloud-edge-end collaboration in the previous embodiment. Figure 2 As shown, this application provides an intelligent computing power integrated service management platform based on cloud-edge-end collaboration. The platform and method embodiments in the embodiments of this application are based on the same inventive concept. Among them, the platform includes:

[0068] The resource topology map building module 10 is used to perceive the computing resource status of the edge collaborative port and the terminal device in real time in the cloud controller and build a resource topology map.

[0069] The dynamic task demand introduction module 20 is used to introduce dynamic task demands based on the resource topology map and generate a computing power scheduling vector including computing power load balancing weights, delay sensitivity indicators and energy consumption constraints.

[0070] The regional elastic cluster composition module 30 is used to perform lightweight segmentation on the cloud training model corresponding to the cloud controller according to the computing power scheduling vector, dynamically match the edge node reasoning capability, and determine the adjacent edge nodes under the hierarchical limitation to form a regional elastic cluster.

[0071] The computing power integration service management module 40 is used to deploy a digital twin simulation engine to pre-run computing power allocation scenarios, dynamically correct resource fragmentation rates using reinforcement learning, establish a heterogeneous resource pooling mechanism, dynamically divide heterogeneous computing power resources within regional elastic clusters on demand, and perform computing power integration service management.

[0072] Furthermore, the platform is also used to implement the following functions:

[0073] Integrate user electricity consumption behavior time series data and equipment infrared thermal imaging data, and extract cross-dimensional features through the spatiotemporal attention mechanism; aggregate the weights of multi-region load forecasting models based on the federated learning framework, dynamically optimize the harmonic distortion rate compensation parameters based on the cross-dimensional features, and compress the cloud-based model parameters through a lightweight knowledge distillation architecture.

[0074] Furthermore, the platform is also used to implement the following functions:

[0075] A dynamic reactive power compensation strategy is set up to generate real-time compensation instructions based on the characteristics of inductive and capacitive loads. The strategy is then combined with LSTM to predict grid voltage fluctuation trends. Hardware-accelerated edge nodes are deployed to process harmonic suppression parameters in parallel, ensuring that the dynamic compensation response time meets response time constraints.

[0076] Furthermore, the platform is also used to implement the following functions:

[0077] The load forecast weight is determined by connecting to the user behavior portrait library, and the partition forecast benchmark curve is formulated in combination with the equipment topology data; based on the partition forecast benchmark curve, a blockchain evidence storage unit is integrated, and the blockchain evidence storage unit is used to record the update log of the prediction model parameters.

[0078] Furthermore, the platform is also used to implement the following functions:

[0079] Based on the computing power allocation scenario, the network bandwidth threshold, latency tolerance upper limit, and energy consumption increment limit under the load transfer path are inserted; and a multi-objective optimization solution is performed with load balancing, transfer cost, and service quality as the goals.

[0080] Furthermore, the platform is also used to implement the following functions:

[0081] Based on the generative adversarial network, the sensor noise interference pattern is simulated and the load feature extraction threshold is dynamically optimized. The digital twin simulation engine is used to correlate the equipment failure history data with the load feature extraction threshold for risk assessment.

[0082] Furthermore, the platform is also used to implement the following functions:

[0083] Based on Monte Carlo simulation, the confidence interval of the partition prediction benchmark curve is dynamically revised; the revised confidence interval is combined with the equipment operation data and compared with the preset risk threshold. If it is exceeded, the early warning mechanism is triggered, and the risk assessment results are fed back to the cloud controller.

[0084] Furthermore, the platform is also used to implement the following functions:

[0085] Based on the energy efficiency ratio of the equipment, combined with photovoltaic output and task processing progress, the distribution ratio of tasks on each edge node is dynamically adjusted; at the same time, the energy consumption of each edge node after computing power is used is monitored, and the heterogeneous resource pooling mechanism is updated.

[0086] Furthermore, the platform is also used to implement the following functions:

[0087] By integrating hardware acceleration cards with containerization, it supports dynamic on-demand division of GPU computing power, CPU computing power, and FPGA computing power. At the same time, it quantitatively evaluates the energy efficiency of GPU computing power, CPU computing power, and FPGA computing power based on energy consumption data, and optimizes the ratio in the regional elastic cluster.

[0088] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0089] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

[0090] This specification and drawings are merely illustrative of the present application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Obviously, those skilled in the art may make various modifications and variations to this application without departing from the scope of this application. Thus, this application is intended to include such modifications and variations as fall within the scope of this application and its equivalents.

Claims

1. An intelligent computing power integrated service management method based on cloud-edge-end collaboration is characterized by: The method comprises: The cloud controller perceives the computing resource status of edge collaborative ports and terminal devices in real time and builds a resource topology map. Based on the resource topology map, dynamic task requirements are introduced to generate a computing power scheduling vector including computing power load balancing weights, delay sensitivity indicators, and energy consumption constraints; Based on the computing power scheduling vector, the cloud training model corresponding to the cloud controller is lightweight segmented, the edge node reasoning capability is dynamically matched, and the adjacent edge nodes under the hierarchical limit are determined to form a regional elastic cluster; Deploy a digital twin simulation engine to rehearse computing power allocation scenarios, combine reinforcement learning to dynamically correct resource fragmentation rates, establish a heterogeneous resource pooling mechanism, dynamically divide heterogeneous computing power resources within regional elastic clusters on demand, and perform computing power integration service management.

2. The intelligent computing power integrated service management method based on cloud-edge-end collaboration according to claim 1 is characterized in that: Performing lightweight segmentation on the cloud training model corresponding to the cloud controller, the method further comprising: Integrate user electricity usage behavior time series data and device infrared thermal imaging data, and extract cross-dimensional features through the spatiotemporal attention mechanism; Based on the federated learning framework, the weights of multi-region load forecasting models are aggregated, and the harmonic distortion rate compensation parameters are dynamically optimized in combination with the cross-dimensional features. The cloud-based model parameters are compressed through a lightweight knowledge distillation architecture.

3. The intelligent computing power integrated service management method based on cloud-edge-end collaboration according to claim 2 is characterized in that: The method further comprises: Set up a dynamic reactive power compensation strategy, generate real-time compensation instructions based on inductive and capacitive load characteristics, and combine it with LSTM to predict grid voltage fluctuation trends; At the same time, hardware-accelerated edge nodes are deployed to process harmonic suppression parameters in parallel, so that the dynamic compensation response time meets the response time limit.

4. The intelligent computing power integrated service management method based on cloud-edge-end collaboration according to claim 1 is characterized in that: The method further comprises: Connect to the user behavior profile library to determine the load forecast weight, and combine the equipment topology data to formulate the partition forecast benchmark curve; Based on the partition prediction benchmark curve, a blockchain evidence storage unit is integrated, and the blockchain evidence storage unit is used to record the prediction model parameter update log.

5. The intelligent computing power integrated service management method based on cloud-edge-end collaboration according to claim 4 is characterized in that: Deploy a digital twin simulation engine to simulate computing power allocation scenarios, combine reinforcement learning to dynamically correct resource fragmentation rates, and establish a heterogeneous resource pooling mechanism. The method includes: Based on the computing power allocation scenario, insert the network bandwidth threshold, latency tolerance upper limit, and energy consumption increment limit under the load transfer path; Multi-objective optimization is performed with load balance, transfer cost and service quality as the goals.

6. The intelligent computing power integrated service management method based on cloud-edge-end collaboration according to claim 5 is characterized in that: The method further comprises: Based on the generative adversarial network, the sensor noise interference pattern is simulated to dynamically optimize the load feature extraction threshold. Use the digital twin simulation engine to correlate equipment failure history data with load feature extraction thresholds for risk assessment.

7. The intelligent computing power integrated service management method based on cloud-edge-end collaboration according to claim 6 is characterized in that: Correlating equipment failure history data with load feature extraction thresholds for risk assessment, the method comprising: Based on Monte Carlo simulation, dynamically correct the confidence interval of the partition prediction benchmark curve; The revised confidence interval is combined with the equipment operation data to be compared with the preset risk threshold. If it is exceeded, the early warning mechanism is triggered and the risk assessment result is fed back to the cloud controller.

8. The intelligent computing power integrated service management method based on cloud-edge-end collaboration according to claim 1 is characterized in that: A heterogeneous resource pooling mechanism is established to dynamically divide heterogeneous computing resources within the regional elastic cluster on demand and perform computing power integration service management. The method further includes: Based on the energy efficiency ratio of the equipment, combined with the photovoltaic output and task processing progress, the distribution ratio of tasks on each edge node is dynamically adjusted; At the same time, the energy consumption of each edge node after computing power is used is monitored, and the heterogeneous resource pooling mechanism is updated.

9. The intelligent computing power integrated service management method based on cloud-edge-end collaboration according to claim 8 is characterized in that: Monitoring the energy consumption of each edge node after computing power is used and updating the heterogeneous resource pooling mechanism, the method includes: Through the integration of hardware acceleration cards and containerization, it supports dynamic on-demand division of GPU computing power, CPU computing power, and FPGA computing power; At the same time, based on the energy consumption data, the energy efficiency of GPU computing power, CPU computing power, and FPGA computing power is quantitatively evaluated, and the ratio is optimized in the regional elastic cluster.

10. An intelligent computing power integrated service management platform based on cloud-edge-end collaboration, characterized by: The platform is used to implement the intelligent computing power integrated service management method based on cloud-edge-end collaboration according to any one of claims 1 to 9, and the platform includes: The resource topology map construction module is used to perceive the computing resource status of edge collaborative ports and terminal devices in real time on the cloud controller and build a resource topology map; A dynamic task demand introduction module is used to introduce dynamic task demands based on the resource topology map and generate a computing power scheduling vector including computing power load balancing weights, delay sensitivity indicators and energy consumption constraints; A regional elastic cluster composition module is used to perform lightweight segmentation on the cloud training model corresponding to the cloud controller according to the computing power scheduling vector, dynamically match the edge node reasoning capabilities, and determine the adjacent edge nodes within the hierarchical limit to form a regional elastic cluster; The computing power integration service management module is used to deploy digital twin simulation engines to rehearse computing power allocation scenarios, dynamically correct resource fragmentation rates using reinforcement learning, establish a heterogeneous resource pooling mechanism, and dynamically divide heterogeneous computing power resources within regional elastic clusters on demand to manage computing power integration services.

Images