Cloud environment elastic capacity expansion arrangement method, system and equipment based on prediction driving and state perception, and medium

Abstract

The invention relates to the technical field of cloud computing resource management, and discloses a cloud environment elastic capacity expansion arrangement method, system and device based on prediction driving and state awareness and a medium, and the method comprises the steps: collecting real-time monitoring data, carrying out the preprocessing, generating a system state representation vector, and storing the system state representation vector as a state sequence; predicting a future load level by adopting a time sequence prediction model, calculating a load growth rate, comparing the load growth rate with a dynamic threshold value, and generating a capacity expansion trigger signal according to a comparison result; constructing a multi-objective optimization model and solving an optimal capacity expansion strategy; executing an automatic expansion operation through the container arrangement platform, and monitoring an expansion completion state in real time; and after capacity expansion is completed, capacity expansion income is calculated, strategy parameters are adjusted according to the capacity expansion income, the updated strategy parameters are fed back to the multi-objective optimization model, and a closed-loop optimization process is formed. According to the method, intelligent upgrading of cloud environment resource scheduling is realized by constructing an elastic capacity expansion arrangement mechanism combining prediction driving and state awareness.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cloud computing resource management, and in particular to a cloud environment elastic scaling arrangement method, system, device and medium based on prediction driving and state perception. BACKGROUND

[0002] Currently, in the field of cloud computing resource management and elastic scaling, there are many technologies related to load monitoring, automatic scaling and resource arrangement. For example, mainstream cloud platforms usually implement dynamic scaling of resources based on monitoring indicators, and can complete instance expansion and traffic distribution through service discovery, load balancing and other mechanisms at the container arrangement level. Some research works introduce prediction algorithms to identify load trends in advance, so as to realize more forward-looking scaling trigger strategies. Some existing technologies combine strategy engines and scripted operation and maintenance means to automatically control the allocation and task arrangement of cloud resources.

[0003] However, the existing technologies still have obvious limitations in actual large-scale cloud environments. First, most existing systems focus on passive monitoring and threshold triggering, lack predictive modeling capabilities based on historical load, and are difficult to respond to sudden loads or periodic fluctuations in a timely manner. Traditional scaling mechanisms often perform resource scheduling only after system performance has declined, resulting in response delays, request accumulation and even service interruptions in the short term. Secondly, current elastic scaling methods mostly rely on a single performance indicator for decision-making, lacking multi-dimensional state perception and global optimization capabilities. For example, existing systems only focus on CPU or memory utilization, ignoring network bandwidth, disk I / O, response latency and inter-service dependencies, leading to inconsistencies between scaling decisions and actual bottlenecks. Furthermore, existing scaling and arrangement strategies mostly rely on static thresholds and rules set by humans, lacking adaptive and feedback optimization mechanisms. The system cannot automatically correct scaling parameters based on historical behavior or environmental changes during long-term operation, leading to model aging and strategy rigidity. When business patterns, hardware performance or load patterns change, the system still follows old rules, and the scaling effect is disconnected from the actual running state. At the same time, the scaling execution results are not fed back to the scheduling engine, and the decision-making lacks closed-loop feedback, making it difficult to achieve continuous evolution and intelligent optimization of strategies.

[0004] Therefore, there is an urgent need for an intelligent elastic scaling arrangement method that combines prediction driving mechanisms and multi-dimensional state perception models, which can not only perceive load change trends in advance to realize proactive decision-making and pre-scaling, but also achieve cross-layer coordinated scheduling through global state modeling. SUMMARY

[0005] In view of the above existing problems, the present application is proposed.

[0006] Therefore, the application provides a cloud environment elastic expansion arrangement method, system, device and medium based on prediction driving and state perception, solves the problem that the existing cloud environment elastic expansion technology mainly depends on a fixed threshold triggering mechanism, lacks predictive modeling ability based on historical load and multi-dimensional state perception, and causes resource scheduling lag, one-sided expansion decision and strategy solidification.

[0007] To solve the above technical problems, the application provides the following technical solutions. In a first aspect, the application provides a cloud environment elastic expansion arrangement method based on prediction driving and state perception, comprising: Collecting real-time monitoring data in multiple dimensions in the cloud environment and preprocessing to generate system state representation vectors and store as state sequences; Based on the state sequence, a time series prediction model is used to predict future load levels, calculate load growth rate and compare with dynamic threshold, and generate expansion triggering signal according to the comparison result; Based on the expansion triggering signal, a multi-objective optimization model is constructed and the optimal expansion strategy is solved; According to the optimal expansion strategy, an automatic expansion operation is performed through a container arrangement platform, and the expansion completion state is monitored in real time; After expansion is completed, the expansion benefit is calculated, and the strategy parameters are adjusted according to the expansion benefit, and the updated strategy parameters are fed back to the multi-objective optimization model to form a closed-loop optimization process.

[0008] As a preferred scheme of the cloud environment elastic expansion arrangement method based on prediction driving and state perception, the generation of the expansion triggering signal comprises: Extracting state representation vectors of the last several time steps from the state sequence, inputting into a time series prediction model to learn the time series dependence in the state sequence, and outputting predicted load values in a specified time window in the future; According to the current load value and the predicted load value, the load growth rate is calculated; Obtain the statistical characteristics of the historical load growth rate, and dynamically generate a triggering threshold based on the statistical characteristics and a preset sensitivity coefficient; The load growth rate is compared with the triggering threshold, if the load growth rate exceeds the triggering threshold, the expansion triggering signal is generated, otherwise it is not triggered.

[0009] The beneficial effects of the preferred technical scheme are that the load change trend can be identified in advance, the expansion signal is triggered in advance, and the problem of response lag of the traditional threshold triggering mechanism is effectively solved.

[0010] As a preferred scheme of the cloud environment elastic expansion arrangement method based on prediction driving and state perception provided by the application, wherein the automatic expansion operation performed by the container arrangement platform according to the optimal expansion strategy comprises: The optimal expansion strategy is received, and the target instance type, quantity and allocation node are parsed out; A resource allocation interface is called through the container arrangement platform, and new cloud hosts or container instances are created according to the parsing result, and in the instance creation process, the startup state and health check result of the instance are monitored in real time; According to the number of instances that have been successfully started and passed the health check, the completion rate of the current expansion operation is calculated; If the completion rate reaches the preset target, it is determined that the expansion operation is successfully completed, otherwise the expansion exception handling process is started.

[0011] As a preferred scheme of the cloud environment elastic expansion arrangement method based on prediction driving and state perception provided by the application, wherein the closed loop optimization process comprises: After the expansion operation is completed, the performance index data before and after the expansion of the system is collected, and the expansion benefit of this expansion operation is calculated based on the performance index data; According to the expansion benefit, the effectiveness of the optimal expansion strategy is judged, and if the expansion effect does not reach the expectation, the sensitivity coefficient of the prediction model or the weight parameter in the multi-objective optimization model is automatically adjusted; The adjusted strategy parameters are fed back to the next round of expansion decision process, and are used to update the solution condition of the multi-objective optimization model, so that the continuous evolution and self-optimization of the strategy are realized.

[0012] The beneficial effects of the preferred technical scheme are that the self-adaptive ability, overall utilization rate and running stability of cloud environment resource management are significantly improved.

[0013] As a preferred scheme of the cloud environment elastic expansion arrangement method based on prediction driving and state perception provided by the application, wherein the system state representation vector comprises: Real-time monitoring data from cloud hosts, containers, networks and storage layers are collected, and the real-time monitoring data includes CPU utilization, memory occupancy, disk I / O throughput, network bandwidth utilization, average response delay and request queue length; The real-time monitoring data is standardized to eliminate dimensional differences; The weighted fusion of each monitoring data after standardization is performed according to a preset weight matrix, and a system state representation vector reflecting the overall running state is generated.

[0014] As a preferred scheme of the cloud environment elastic capacity expansion arrangement method based on prediction driving and state perception provided by the application, wherein the constructing a multi-objective optimization model comprises: Taking system performance improvement, resource use cost and energy consumption as optimization objectives, a multi-objective optimization function is established; weight coefficients of each optimization objective in the multi-objective optimization function are determined, and the weight coefficients reflect the importance of different optimization objectives in decision-making; Lower limit constraints of the system performance improvement, upper limit constraints of the resource use cost and upper limit constraints of the energy consumption are set.

[0015] As a preferred scheme of the cloud environment elastic capacity expansion arrangement method based on prediction driving and state perception provided by the application, wherein the solving an optimal expansion strategy comprises: According to the current system state and the scalable node type, a group of feasible expansion schemes are randomly generated as an initial candidate solution set; For each expansion scheme in the candidate solution set, the objective function value under the multi-objective optimization function is calculated, and whether all preset constraint conditions are met is verified; According to the objective function value, the candidate solution set is selected and operated, and the expansion scheme that meets the constraint condition and has a higher objective function value is retained; The selected expansion scheme is executed with a crossover operation, new expansion schemes are generated by combining parameters of different schemes, and the expansion scheme after the crossover is executed with a mutation operation, and the node type or quantity parameter is randomly adjusted to explore a more optimal solution space; The above calculation, selection, crossover and mutation steps are repeated until a preset termination condition is reached, and the expansion scheme with the optimal objective function value is output as the optimal expansion strategy.

[0016] The preferred technical scheme has the beneficial effects that the optimal resource configuration under the system performance requirement is realized, and the limitation of single index decision-making is overcome.

[0017] In a second aspect, the application provides a cloud environment elastic capacity expansion arrangement system based on prediction driving and state perception, comprising: A data processing module is configured to collect real-time monitoring data in multiple dimensions in a cloud environment, pre-process the data, generate a system state representation vector, and store the vector as a state sequence; A load trend analysis module is configured to predict future load levels based on the state sequence using a time series prediction model, calculate a load growth rate, compare the load growth rate with a dynamic threshold, and generate an expansion trigger signal according to the comparison result; A dynamic expansion decision and optimization module is configured to construct a multi-objective optimization model and solve an optimal expansion strategy based on the expansion trigger signal; The intelligent orchestration and execution control module is used to perform automatic scaling operations through the container orchestration platform according to the optimal scaling strategy and monitor the scaling completion status in real time. The self-feedback and strategy evolution module is used to calculate the expansion benefits after the expansion is completed, adjust the strategy parameters according to the expansion benefits, and feed the updated strategy parameters back to the multi-objective optimization model to form a closed-loop optimization process.

[0018] Thirdly, the present invention provides an electronic device, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor executes the computer-executable instructions to implement the steps of a cloud environment elastic scaling and orchestration method based on prediction-driven and state-aware principles.

[0019] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement steps of a predictive-driven and state-aware cloud environment elastic scaling orchestration method.

[0020] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention achieves an intelligent upgrade of cloud environment resource scheduling by constructing a flexible expansion orchestration mechanism that combines prediction-driven and state-aware approaches. Based on real-time acquisition and state modeling of multi-dimensional monitoring data, combined with time series prediction technology, this invention can proactively identify load change trends and trigger expansion signals in advance, effectively solving the problem of delayed response in traditional threshold triggering mechanisms. Secondly, by establishing a multi-objective optimization model that integrates performance, cost, and energy consumption, and employing intelligent optimization algorithms to solve for the optimal expansion strategy, this invention achieves optimal resource allocation while meeting system performance requirements, overcoming the limitations of single-index decision-making. Furthermore, through self-feedback and strategy evolution mechanisms, the system can dynamically adjust decision parameters based on the expansion execution effect, forming a continuously optimized closed-loop control, significantly improving the adaptive capability, overall utilization rate, and operational stability of cloud environment resource management. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the overall process logic of a cloud environment elastic scaling and orchestration method based on prediction-driven and state-awareness, provided as an embodiment of the present invention. Detailed Implementation

[0023] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present application.

[0024] Embodiment 1, refer to Figure 1 For an embodiment of the present application, a cloud environment elastic expansion arrangement method based on prediction driving and state perception is provided, as shown in Figure 1 Specifically includes the following steps: S100: Collecting real-time monitoring data in multiple dimensions in the cloud environment and preprocessing, generating system state feature vector, and storing as state sequence; S200: Based on the state sequence, a time series prediction model is used to predict the future load level, the load growth rate is calculated and compared with the dynamic threshold, and the expansion trigger signal is generated according to the comparison result; S300: Based on the expansion trigger signal, a multi-objective optimization model is constructed and the optimal expansion strategy is solved; S400: According to the optimal expansion strategy, the automatic expansion operation is executed through the container arrangement platform, and the expansion completion state is monitored in real time; S500: Calculate the expansion benefit after the expansion is completed, and adjust the strategy parameters according to the expansion benefit, and feed back the updated strategy parameters to the multi-objective optimization model to form a closed-loop optimization process.

[0025] It should be noted that, in order to solve the problem that the existing cloud environment elastic expansion technology mainly relies on fixed threshold trigger mechanism, lacks predictive modeling ability based on historical load and multi-dimensional state perception, resulting in resource scheduling lag, expansion decision-making one-sided and strategy solidification, the above steps S100~S500 construct an elastic expansion arrangement mechanism combining prediction driving and state perception, realizing the intelligent upgrading of cloud environment resource scheduling. Based on the real-time collection and state modeling of multi-dimensional monitoring data, combined with time series prediction technology, the present application can prospectively identify the load change trend, trigger the expansion signal in advance, and effectively solve the problem of response lag of traditional threshold trigger mechanism. Secondly, the present application establishes a multi-objective optimization model integrating performance, cost and energy consumption, and solves the optimal expansion strategy by using intelligent optimization algorithm, realizes the optimal configuration of resources under the condition of meeting the system performance requirements, and overcomes the limitations of single index decision. In addition, through the self-feedback and strategy evolution mechanism, the system can dynamically adjust the decision parameters according to the expansion execution effect, form a continuous optimization closed loop control, and significantly improve the adaptive ability, overall utilization rate and operation stability of cloud environment resource management.

[0026] In the embodiment of the present application, the step S100 of collecting and preprocessing the real-time monitoring data of multiple dimensions in the cloud environment, generating a system state representation vector, and storing it as a state sequence includes: Specifically, real-time monitoring data from cloud hosts, containers, network and storage layers are collected, wherein the real-time monitoring data includes CPU utilization, memory occupancy, disk I / O throughput, network bandwidth utilization, average response delay and request queue length, and the formula is expressed as: wherein, represents a set of real-time monitoring data, represents CPU utilization, represents memory occupancy, represents disk I / O throughput, represents network bandwidth utilization, represents average response delay, represents request queue length.

[0027] Specifically, the real-time monitoring data is standardized to eliminate dimensional differences, and the weighted fusion of each monitoring data after standardization is performed according to a preset weight matrix to generate a system state representation vector reflecting the overall running state , and the formula is expressed as: wherein, is a weight matrix used to describe the influence degree of different indicators on the overall health degree, and are the historical mean and standard deviation, respectively.

[0028] It should be noted that the system state representation vector reflecting the overall running state is stored as a state sequence for subsequent trend prediction and decision modeling.

[0029] It should be noted that the step S100 collects the monitoring data of multiple dimensions in the cloud environment in real time, and performs standardization and weighted fusion processing to generate a state representation vector that can fully reflect the system running state, breaking through the limitations of traditional single indicator monitoring, and building a unified multi-dimensional state perception system to improve the integrity and accuracy of state perception from the source.

[0030] In the embodiment of the present application, the step S200 of predicting the future load level based on the state sequence using a time series prediction model, calculating the load growth rate and comparing it with a dynamic threshold, and generating an expansion trigger signal according to the comparison result includes the following sub-steps B1-B4: In B1: the state representation vectors of the recent time steps are extracted from the state sequence, input into the time series prediction model to learn the time series dependency in the state sequence, and output the predicted load values of the future specified time window; Specifically, the state representation vectors of the recent time steps are extracted from the state sequence, input into the time series prediction model, and the time series prediction model is used to model the state sequence and predict the future load level , which is expressed by the formula: It should be noted that the processing mechanism of the time series prediction model is to collect the state vectors of the recent n time steps , including load, resource occupancy and other features related to system state. The sequence is input into the time series model , and the time series dependency is learned by the model. The model gives the predicted value of the future step according to the historical sequence , which is the growth rate of t steps.

[0031] In B2: the load growth rate is calculated according to the current load value and the predicted load value , which is expressed by the formula: In B3: the statistical features of the historical load growth rate are obtained, and the trigger threshold is dynamically generated based on the statistical features and the preset sensitivity coefficient; Specifically, the system dynamically generates the trigger threshold by statistically analyzing the historical growth rate distribution : wherein is the mean of the historical growth rate, is the sensitivity coefficient, which is generally 0.5, is the standard deviation of the growth rate.

[0032] In B4: the load growth rate is compared with the trigger threshold, if the load growth rate exceeds the trigger threshold, i.e. , the expansion trigger signal is generated , otherwise it is not triggered .

[0033] In an optional embodiment, the processing of the time series prediction model can also be based on the prediction method of the time series convolution network, which uses layer stacking of dilated causal convolution layers to capture multi-scale time series patterns in long-term historical state sequences, and effectively models long-range dependencies by expanding the receptive field.

[0034] In another optional embodiment, the processing of the time series prediction model can also be based on a prediction method of the Transformer, dynamically calculating the correlation weight of different time steps in the state sequence through a self-attention mechanism, and processing global time sequence information in parallel, which is particularly suitable for capturing the load variation law with complex periodicity and mutation characteristics.

[0035] It should be noted that the above step S200 realizes the transition from passive response to active prediction, can identify the load growth trend in advance, triggers the capacity expansion preparation before the system performance declines, and effectively avoids the problems of response delay or service interruption caused by resource scheduling lag.

[0036] In the embodiment of the application, the step S300 of constructing a multi-objective optimization model and solving an optimal capacity expansion strategy based on the capacity expansion trigger signal comprises the following sub-steps C1 and C2: In C1: a multi-objective optimization model is constructed; Specifically, the system performance improvement, the resource use cost and the energy consumption are taken as the optimization objectives, a multi-objective optimization function is established, the weight coefficients of the optimization objectives in the multi-objective optimization function are determined, and the weight coefficients reflect the importance of different optimization objectives in decision-making; the formula is: Among them, represents the capacity expansion scheme variable (including the number of instances, the resource type, etc.), represents a capacity expansion cost function, represents an energy consumption cost function, represents a performance improvement function, all represent weight coefficients, and satisfy .

[0037] Specifically, the lower limit constraint of the system performance improvement, the upper limit constraint of the resource use cost and the upper limit constraint of the energy consumption are set, and the formula is: Among them, the performance lower limit is generally 0.95 times the performance, the cost upper limit is generally 1.3 times the capacity expansion cost, the energy consumption upper limit is generally 1.2 times the basic energy consumption, and the specific value depends on the system settings.

[0038] In an optional embodiment, the multi-objective optimization model can also be a multi-objective optimization model based on the Pareto optimal theory, which maintains a non-dominated solution set and uses crowding degree calculation to maintain the diversity of the solution set, and directly finds a Pareto optimal solution set that satisfies multiple constraints.

[0039] In another optional embodiment, the multi-objective optimization model can also be an optimization model using the weighted Chebyshev method, which effectively handles the balanced optimization problem between different dimensional objectives by converting the multi-objective into a weighted maximum distance between the minimum objective function and the ideal solution.

[0040] In C2: solving the optimal expansion strategy; the detailed steps include: According to the current system state and the expandable node types, a set of feasible expansion schemes are randomly generated as the initial candidate solution set where each individual represents an expansion decision, such as "expand 2 nodes, type A + type B"; For each expansion scheme in the candidate solution set, calculate the objective function value under the multi-objective optimization function, and verify whether it meets all the preset constraint conditions; According to the objective function value, the candidate solution set is selected and operated, and the expansion scheme that meets the constraint condition and has a higher objective function value is retained; Perform crossover operation on the selected expansion scheme, combine the parameters of different schemes to generate new expansion schemes, and perform mutation operation on the expansion schemes after crossover to randomly adjust the node type or quantity parameters to explore better solution space; Repeat the above calculation, selection, crossover and mutation steps until the preset termination condition is reached, output the expansion scheme with the optimal objective function value as the optimal expansion strategy, which is represented by the formula: It should be noted that the preset constraint conditions include that the performance improvement needs to reach the system set lower limit, the resource cost cannot exceed the budget upper limit, and the energy consumption needs to be controlled within the device bearing range; the preset termination condition is to reach the maximum iteration number or the change amplitude of the objective function value of the optimal solution in continuous generations is less than the convergence threshold, ensuring that the algorithm obtains stable optimization solution within reasonable calculation cost.

[0041] In an optional embodiment, the step of solving the optimal expansion strategy can also be based on the decision-making method of reinforcement learning, which evaluates the cumulative returns of different expansion schemes in the long-term operation by constructing a state-action value function, and gradually optimizes the decision-making strategy using experience replay and policy gradient method.

[0042] In another optional embodiment, the step of solving the optimal expansion strategy can also use a multi-objective particle swarm optimization algorithm, which stores non-dominated solutions in an external archive set and performs elite preservation according to the crowding distance, so that the search process quickly converges to the Pareto optimal frontier.

[0043] It should be noted that the above step S300 overcomes the one-sidedness of the traditional single-objective decision by constructing a multi-objective optimization model and solving the optimal expansion strategy by using an intelligent algorithm, and can maximize the overall benefit of the system while meeting the performance requirements, taking into account resource use efficiency and economy.

[0044] In the embodiment of the application, the step S400 described above performs an automatic expansion operation according to the optimal expansion strategy through a container orchestration platform, and real-time monitoring of the expansion completion state includes: Receiving the optimal expansion strategy, the target instance type, number and allocation node are parsed out; Through the resource allocation interface of the container orchestration platform, new cloud hosts or container instances are created according to the parsing results, and in the instance creation process, the startup state and health check results of the instance are monitored in real time; According to the number of instances that have been successfully started and passed the health check, the completion rate of the current expansion operation is calculated; If the completion rate reaches the preset target, it is determined that the expansion operation is successfully completed, otherwise the expansion exception handling process is started.

[0045] Specifically, based on the Kubernetes, OpenStack or Docker Swarm framework, the automatic expansion operation is performed. The system adopts a progressive expansion strategy: Among them, represents the number of allocated instances in the kth round, represents the adjustment parameter for controlling the expansion rate, represents the target expansion number, represents the initial expansion number.

[0046] Specifically, according to the number of instances that have been successfully started and passed the health check, the completion rate of the current expansion operation is calculated, and the formula is: Specifically, when =1, the expansion is completed, and the related running state record enters the feedback stage, Not equal to 1, the system enters the expansion exception handling process.

[0047] In an optional embodiment, the automatic expansion operation can also use a progressive replacement strategy based on blue-green deployment, which creates a complete new version environment in parallel and gradually switches traffic, ensuring service continuity while completing resource expansion.

[0048] In another optional embodiment, the automatic scaling operation can also employ an elastic resource pool pre-allocation mechanism to pre-maintain a hot backup instance resource pool, and directly enable the pre-configured instance and quickly access the service cluster through a dynamic registration mechanism when scaling is triggered.

[0049] It should be noted that the step S400 automatically executes the scaling strategy through the container orchestration platform, and monitors the scaling progress and node state in real time, realizes the automation and visualization of resource allocation, ensures that the scaling operation is accurately landed according to the expected target, and guarantees the reliability and stability of the scaling process through the exception detection and processing mechanism.

[0050] In the embodiment of the application, the step S500 calculates the scaling benefit after the scaling is completed, adjusts the strategy parameters according to the scaling benefit, feeds back the updated strategy parameters to the multi-objective optimization model, and forms a closed-loop optimization process, including: After the scaling operation is completed, the performance index data before and after scaling is collected, and the scaling benefit of this scaling operation is calculated based on the performance index data, which is expressed by the formula: According to the scaling benefit, the effectiveness of the optimal scaling strategy is judged, and if the scaling effect does not reach the expectation, the sensitivity coefficient of the prediction model or the weight parameter in the multi-objective optimization model is automatically adjusted; The adjusted strategy parameters are fed back to the next round of scaling decision process, which is used to update the solution condition of the multi-objective optimization model, so as to realize the continuous evolution and self-optimization of the strategy.

[0051] Specifically, if the scaling effect does not reach the expectation, the prediction sensitivity is enhanced or the decision weight is increased If the resource idle rate rises after scaling, the trigger threshold is reduced or the parameter is adjusted The strategy updating rule is: Wherein, represents a comprehensive benefit function, is a learning rate.

[0052] It should be noted that the step S500 calculates the actual benefit after each scaling is completed, and dynamically adjusts the strategy parameters based on the feedback result to form a closed-loop optimization mechanism, so that the system has the ability of continuous learning and self-evolution, can continuously correct the decision model according to the actual running effect, gradually improves the adaptability and precision of the scaling strategy, and realizes the performance optimization and resource saving in the long-term running.

[0053] Embodiment 2 provides an application example of the cloud environment elastic scaling orchestration method based on prediction driving and state sensing, which verifies and illustrates the technical effects used in the method.

[0054] This embodiment takes an online service as an example to illustrate the execution process of the method of the present application. The current running instances of a certain cloud service cluster are 10, and the key running indicators are shown in Table 1. The prediction window is set to Tp=5 minutes.

[0055] Table 1: Key running indicators.

[0056] In this embodiment, the system collects the state vectors of the last 20 time steps, inputs the LSTM model, and obtains a prediction result of 4300 for the next 5 minutes. According to the growth rate, the trigger threshold is calculated: rt = 0.26 > θp = 0.18 → trigger the expansion signal. The multi-objective optimization model is solved under the constraint, and the optimal expansion scheme is obtained: expand 1 instance (same type of node). This is the scheme with the lowest cost calculated by the GA optimizer, and the performance improvement meets the requirements. The system automatically creates a new node node-11, which completes the startup and health check within 35 seconds. The expansion completion rate is calculated to be 1, so the expansion is determined to be successful. The system performance changes after 1 minute of expansion are shown in Table 2.

[0057] Table 2: System performance changes after 1 minute of expansion.

[0058] From the above results, it can be seen that the present application realizes the intelligent upgrading of cloud environment resource scheduling by constructing an elastic expansion orchestration mechanism combining prediction driving and state perception. Based on real-time collection and state modeling of multi-dimensional monitoring data, combined with time series prediction technology, the present application can prospectively identify the load change trend, trigger the expansion signal in advance, and effectively solve the problem of response lag of the traditional threshold trigger mechanism. Secondly, the present application establishes a multi-objective optimization model integrating performance, cost and energy consumption, and solves the optimal expansion strategy by using an intelligent optimization algorithm, realizes the optimal allocation of resources under the requirement of system performance, and overcomes the limitations of single indicator decision. In addition, through the self-feedback and strategy evolution mechanism, the system can dynamically adjust the decision parameters according to the expansion execution effect, form a closed-loop control of continuous optimization, and significantly improve the adaptive ability, overall utilization rate and running stability of cloud environment resource management.

[0059] In this embodiment, a cloud environment elastic expansion orchestration system based on prediction driving and state perception is provided, which comprises: A data processing module for collecting and preprocessing real-time monitoring data in multiple dimensions of the cloud environment, generating system state representation vectors, and storing them as state sequences; a load trend analysis module configured to predict future load levels based on the state sequence using a time series prediction model, calculate a load growth rate and compare the load growth rate with a dynamic threshold, and generate a scale-out trigger signal according to a result of the comparison; a dynamic scale-out decision and optimization module configured to construct a multi-objective optimization model and solve an optimal scale-out strategy based on the scale-out trigger signal; an intelligent orchestration and execution control module configured to execute an automatic scale-out operation through a container orchestration platform according to the optimal scale-out strategy, and monitor a scale-out completion state in real time; a self-feedback and strategy evolution module configured to calculate a scale-out benefit after the scale-out is completed, adjust strategy parameters according to the scale-out benefit, and feed back the updated strategy parameters to the multi-objective optimization model to form a closed-loop optimization process.

[0060] It should be noted that the technical scheme of the cloud environment elastic scale-out orchestration system based on prediction driving and state sensing and the technical scheme of the cloud environment elastic scale-out orchestration method based on prediction driving and state sensing described above belong to the same concept. The details of the technical scheme of the cloud environment elastic scale-out orchestration system based on prediction driving and state sensing in the present embodiment are not described in detail, and can be referred to the description of the technical scheme of the cloud environment elastic scale-out orchestration method based on prediction driving and state sensing described above.

[0061] The above-mentioned unit modules can be embedded in or independent of the processor in the electronic device in hardware form, or can be stored in the memory in the electronic device in software form, so as to be called and executed by the processor to perform the operations corresponding to the above-mentioned modules.

[0062] The present embodiment also provides an electronic device, which includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the electronic device is configured to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the electronic device is configured to communicate with external terminals in wired or wireless manner. The wireless manner can be achieved through WIFI, operator network, NFC (near field communication) or other technologies. The computer program is executed by the processor to implement the cloud environment elastic scale-out orchestration method based on prediction driving and state sensing. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen. The input device of the electronic device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the electronic device, or an external keyboard, touchpad or mouse, etc.

[0063] The embodiment further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method proposed in the above embodiment.

[0064] The storage medium proposed in the embodiment belongs to the same inventive concept as the method proposed in the above embodiment, and the technical details not described in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0065] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary universal hardware, and of course can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory, a random access memory, a flash memory, a hard disk or an optical disk, etc., and includes a number of instructions to make an electronic device (which can be a personal computer, a server, or a network device, etc.) execute the method of the embodiments of the present application.

[0066] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application rather than limit the present application, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A cloud environment elastic scaling arrangement method based on prediction-driven and state-aware, characterized in that, The method comprises the following steps: Collecting real-time monitoring data in multiple dimensions in a cloud environment and preprocessing the data to generate system state representation vectors and store the vectors as state sequences; Based on the state sequences, a time series prediction model is used to predict future load levels, calculate load growth rates and compare them with dynamic thresholds to generate expansion trigger signals; Based on the expansion trigger signals, a multi-objective optimization model is constructed and the optimal expansion strategy is solved; According to the optimal expansion strategy, automatic expansion operations are performed through a container orchestration platform, and the expansion completion status is monitored in real time; After the expansion is completed, the expansion benefit is calculated, and the strategy parameters are adjusted according to the expansion benefit. The updated strategy parameters are fed back to the multi-objective optimization model to form a closed-loop optimization process. 2.The cloud environment elastic scaling arrangement method based on prediction-driven and state-awareness according to claim 1, wherein, The generation of the expansion trigger signal comprises: Extracting the state representation vectors of the recent time steps from the state sequences, inputting them into the time series prediction model to learn the time series dependence in the state sequences, and outputting the predicted load values in the future specified time window; According to the current load value and the predicted load value, the load growth rate is calculated; Obtain the statistical characteristics of the historical load growth rate, and dynamically generate the trigger threshold based on the statistical characteristics and the preset sensitivity coefficient; Compare the load growth rate with the trigger threshold. If the load growth rate exceeds the trigger threshold, generate an expansion trigger signal, otherwise do not trigger. 3.The cloud environment elastic scaling arrangement method based on prediction-driven and state-awareness according to claim 2, wherein, The automatic expansion operation according to the optimal expansion strategy through the container orchestration platform comprises: Receive the optimal expansion strategy, parse the target instance type, quantity and allocation node; Call the resource allocation interface through the container orchestration platform, create new cloud hosts or container instances according to the parsing results, and monitor the startup status and health check results of the instances in the instance creation process in real time; According to the number of instances that have successfully started and passed the health check, the completion rate of the current expansion operation is calculated; If the completion rate reaches the preset target, it is determined that the expansion operation is successfully completed, otherwise the expansion exception handling process is started. 4.The cloud environment elastic scaling arrangement method based on prediction-driven and state-awareness according to claim 3, wherein, The closed-loop optimization process comprises: After the expansion operation is completed, the performance index data of the system before and after the expansion is collected, and the expansion benefit of this expansion operation is calculated based on the performance index data; According to the expansion benefit, the effectiveness of the optimal expansion strategy is judged. If the expansion effect does not meet the expectation, the sensitivity coefficient of the prediction model or the weight parameter in the multi-objective optimization model is automatically adjusted; The adjusted strategy parameters are fed back to the next round of expansion decision process to update the solving conditions of the multi-objective optimization model, realizing the continuous evolution and self-optimization of the strategy.

5. The cloud environment elastic scaling arrangement method based on prediction driving and state perception according to claim 1, wherein, The generation of the system state representation vector comprises: Collecting real-time monitoring data from cloud hosts, containers, networks and storage layers, including CPU utilization, memory occupancy, disk I / O throughput, network bandwidth utilization, average response delay and request queue length; Standardize the real-time monitoring data to eliminate dimensional differences; According to the preset weight matrix, the standardized monitoring data is weighted and fused to generate a system state representation vector reflecting the overall running state. 6.The cloud environment elastic scaling arrangement method based on prediction-driven and state-awareness according to claim 5, wherein, The construction of the multi-objective optimization model comprises: A multi-objective optimization function is established with system performance improvement, resource usage cost and energy consumption as optimization objectives; weight coefficients of each optimization objective in the multi-objective optimization function are determined, which reflect the importance of different optimization objectives in decision-making; Lower limit constraints of the system performance improvement, upper limit constraints of the resource usage cost and upper limit constraints of the energy consumption are set.

7. The cloud environment elastic scaling arrangement method based on prediction driving and state perception according to claim 6, wherein, The solving of the optimal expansion strategy includes: A set of feasible expansion schemes are randomly generated as an initial candidate solution set according to the current system state and the scalable node types; For each expansion scheme in the candidate solution set, the objective function value under the multi-objective optimization function is calculated, and whether all preset constraint conditions are met is verified; According to the objective function value, the candidate solution set is selected and operated, and the expansion scheme that meets the constraint condition and has a higher objective function value is retained; The selected expansion scheme is executed with a crossover operation, new expansion schemes are generated by combining parameters of different schemes, and the expansion scheme after the crossover is executed with a mutation operation to randomly adjust the node type or quantity parameter to explore a better solution space; The above calculation, selection, crossover and mutation steps are repeated until a preset termination condition is reached, and the expansion scheme with the optimal objective function value is output as the optimal expansion strategy.

8. A cloud environment elastic scaling arrangement system based on prediction driving and state awareness, applying the cloud environment elastic scaling arrangement method based on prediction driving and state awareness as claimed in any one of claims 1-7, characterized in that, It includes: A data processing module is configured to collect real-time monitoring data in multiple dimensions in a cloud environment, pre-process the data, generate a system state representation vector, and store the vector as a state sequence; A load trend analysis module is configured to predict future load levels based on the state sequence using a time series prediction model, calculate a load growth rate, compare the load growth rate with a dynamic threshold, and generate an expansion trigger signal based on the comparison result; A dynamic expansion decision and optimization module is configured to construct a multi-objective optimization model based on the expansion trigger signal and solve an optimal expansion strategy; An intelligent orchestration and execution control module is configured to execute an automatic expansion operation through a container orchestration platform based on the optimal expansion strategy, and monitor the expansion completion state in real time; A self-feedback and strategy evolution module is configured to calculate an expansion benefit after the expansion is completed, adjust strategy parameters based on the expansion benefit, and feed back the updated strategy parameters to the multi-objective optimization model to form a closed-loop optimization process. 9.An electronic device comprising a memory and a processor, the electronic device characterized by: The memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to implement the steps of the cloud environment elastic expansion orchestration method based on prediction driving and state perception according to any one of claims 1-7.

10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: The computer executable instructions are executed by the processor to implement the steps of the cloud environment elastic expansion orchestration method based on prediction driving and state perception according to any one of claims 1-7.

Images