Resource scheduling method in cloud environment

Abstract

The invention relates to the technical field of resource scheduling, in particular to a resource scheduling method in a cloud environment, which comprises the following steps of: based on resource use data in the cloud environment, performing classified statistics on use frequency and duration of each resource, calculating use frequency distribution of differentiated resources, and obtaining a resource use frequency analysis result. According to the method, the hidden Markov model is adopted to dynamically predict the resource use state, the perspectiveness and adaptability of resource scheduling are enhanced, the complex resource scheduling problem is effectively decomposed into manageable sub-problems through the Lagrangian relaxation technology, the problem processing flow is simplified, the solving efficiency and accuracy are improved, and the resource scheduling efficiency is improved. Through independent solution and optimization of each sub-problem, the overall resource scheduling is more flexible and efficient, the service quality and stability are further improved, the dynamic management capability and response speed of resources are greatly improved, and more stable and efficient services are provided for users.

Description

Technical Field

[0001] The present invention relates to the technical field of resource scheduling, and particularly to a resource scheduling method in a cloud environment. Background Art

[0002] The technical field of resource scheduling involves methods and systems for allocating and managing resources in various computing and network environments. The core is to optimize resource utilization, reduce latency, and improve service quality. In a cloud computing environment, resource scheduling is particularly important because it is necessary to dynamically manage and allocate a large number of dispersed resources located in different geographical locations, including computing power, storage space, and network bandwidth. An effective resource scheduling strategy can ensure the maximization of application performance and the improvement of cost efficiency, while meeting the needs of different users and applications.

[0003] Among them, the resource scheduling method in a cloud environment refers to a technology for dynamically allocating cloud resources (such as virtual machines, storage, bandwidth, etc.) through various algorithms and strategies in a cloud computing architecture. Its main purpose is to improve resource utilization, reduce operating costs, while ensuring service quality and system stability. Through an intelligent scheduling system, the resource allocation can be automatically adjusted according to real-time requirements and preset strategies to cope with changing load conditions and user needs.

[0004] Existing resource scheduling technologies face problems of inflexible resource allocation and insufficient prediction in dynamic and large-scale cloud environments. Especially when dealing with peak and valley resource demands, the future resource usage status cannot be effectively predicted, resulting in insufficient response speed and accuracy of resource allocation. Static or overly history-data-dependent scheduling strategies are difficult to adapt to rapidly changing load demands in practical applications, causing resource waste or shortages. For example, when the load peak cannot be accurately predicted, it leads to insufficient computing resources, affecting users' business operations, while during periods of low resource demand, excessive resource allocation increases operating costs. Existing scheduling methods lack the ability to effectively decompose complex problems, limiting the overall resource management efficiency and being unable to handle large-scale scheduling problems effectively, making it difficult to meet the diverse and real-time changing service demands in a cloud environment. Summary of the Invention

[0005] The purpose of the present invention is to solve the drawbacks existing in the prior art, and to propose a resource scheduling method in a cloud environment.

[0006] To achieve the above purpose, the present invention adopts the following technical solution: A resource scheduling method in a cloud environment, including the following steps:

[0007] S1: Based on the resource usage data in the cloud environment, classify and statistically analyze the usage frequency and duration of each resource, calculate the usage frequency distribution of differentiated resources, and obtain the resource usage frequency analysis result;

[0008] S2: According to the resource usage frequency analysis result, set the initial state probability matrix and the state transition probability matrix of the hidden Markov model, iteratively optimize the model parameters, and obtain the hidden state model parameters;

[0009] S3: Use the hidden state model parameters to predict the state of resource usage within the next five years through the cloud environment, calculate the probability of state transition, predict the state of resource demand, and obtain the resource demand prediction result;

[0010] S4: Through the resource demand prediction result, decompose the resource scheduling problem into multiple sub-problems by the Lagrangian relaxation technique, set the resource type, demand, and constraint conditions for the sub-problems, and obtain the sub-problem decomposition result;

[0011] S5: Utilize the sub-problem decomposition result to adjust the constraint conditions of the sub-problems, iteratively solve the optimal solutions of the sub-problems, and adjust the Lagrangian multipliers to obtain the Lagrangian multiplier optimization result;

[0012] S6: According to the Lagrangian multiplier optimization result, perform dynamic allocation of resources for the predicted resource demand peaks and valleys, formulate and implement the resource allocation execution plan, and obtain the resource scheduling execution strategy.

[0013] As a further solution of the present invention, the resource usage frequency analysis result includes the average usage time, time period, and usage frequency distribution of each type of resource, the hidden state model parameters include the state transition probability matrix, emission probability matrix, and initial probability of the hidden state optimized according to the resource usage data, the resource demand prediction result includes the probability prediction of the peaks and valleys of resource usage in time periods within the next five years and the corresponding resource demand quantities, the sub-problem decomposition result includes the resource demand types, quantitative demand indicators, and corresponding constraint conditions of the sub-problems, the Lagrangian multiplier optimization result includes the adjusted constraint condition weights, iteration times, and convergence states, and the resource scheduling execution strategy includes calculating the resource allocation ratio, adjusting the network bandwidth priority, and dynamically allocating the storage space.

[0014] As a further solution of the present invention, based on the resource usage data in the cloud environment, the steps of classifying and statistically calculating the usage frequency and duration of each resource, and calculating the usage frequency distribution of differentiated resources to obtain the resource usage frequency analysis result are specifically as follows:

[0015] S101: Extract the usage data of differentiated resources from the cloud environment, record the access time point and end time point of each resource, calculate the continuous usage duration of each resource, and generate the resource usage record;

[0016] S102: Based on the resource usage records, screen and organize the data, group the resources according to the usage frequency and duration, and mark each group of data to distinguish different resource types, generating a resource classification result;

[0017] S103: According to the resource classification result, count the usage frequency of each resource type, calculate the usage of each type of resource in different time periods, and organize them by category, generating a resource usage frequency analysis result.

[0018] As a further solution of the present invention, the steps of setting the initial state probability matrix and state transition probability matrix of the hidden Markov model and iteratively optimizing the model parameters according to the resource usage frequency analysis result to obtain the hidden state model parameters are specifically as follows:

[0019] S201: Using the resource usage frequency analysis result and the entropy weight method, allocate weights to the initial probabilities of the hidden states of the hidden Markov model, generating an initial state probability matrix;

[0020] S202: Using the initial state probability matrix and the resource usage frequency data, calculate the probability of resource state conversion, set and adjust the state transition probability matrix, and optimize the transition probability, generating a state transition probability matrix;

[0021] S203: Based on the state transition probability matrix, iteratively calculate to optimize the model parameters, and adjust the parameters of each iteration to avoid prediction errors, determine the optimal values of the model parameters, and generate the hidden state model parameters.

[0022] As a further solution of the present invention, the formula of the entropy weight method is as follows:

[0023]

[0024] where P i is the initial state probability of each resource, p ij represents the usage frequency of the i-th type of resource in the j-th time period, n j represents the total number of time periods, m i represents the total number of resource categories, k represents the adjustment coefficient of information entropy, and e is the base of the natural logarithm.

[0025] As a further solution of the present invention, the steps of using the hidden state model parameters to predict the resource usage status in the next five years through the cloud environment, calculating the probability of state transition, and predicting the resource demand status to obtain the resource demand prediction result are specifically as follows:

[0026] S301: Initialize the prediction model of resource usage status through the hidden state model parameters, configure the model to simulate the resource usage scenarios in the next five years, and generate the model prediction scenarios.

[0027] S302: According to the model prediction scenarios, use the hidden Markov model to calculate the state transition probabilities, predict the probabilities from the real-time state to the states in the next five years, and iteratively update the state probabilities to generate the state transition probability prediction results.

[0028] S303: Based on the state transition probability prediction results, analyze and calculate the state changes of resource requirements, predict the usage requirements and supply situations of various resources, and organize the prediction information of resource requirements in the next five years to generate the resource requirement prediction results.

[0029] As a further solution of the present invention, through the resource requirement prediction results, the resource scheduling problem is decomposed into multiple sub-problems by the Lagrangian relaxation technique. The steps of setting the resource type, requirement, and constraint conditions for the sub-problems to obtain the sub-problem decomposition results are specifically as follows:

[0030] S401: Utilize the resource requirement prediction results, use the Lagrangian relaxation technique to decompose the resource scheduling problem into multiple independent sub-problems, determine the resource type and requirement for each sub-problem, and set the initial constraint conditions to generate the sub-problem initial settings.

[0031] S402: Adopt the sub-problem initial settings, define the resource requirements and constraint information of each sub-problem, adjust and optimize the constraint conditions of the sub-problems, and match the target situations of different resource types and requirements to generate the sub-problem constraint optimization results.

[0032] S403: According to the sub-problem constraint optimization results, perform resource scheduling and requirement matching for each sub-problem, check the optimal allocation and usage efficiency of resources, and obtain the resource allocation plan for the sub-problems to generate the sub-problem decomposition results.

[0033] As a further solution of the present invention, the formula of the Lagrangian relaxation technique is as follows:

[0034]

[0035] Among them, R z represents the resource allocation result of the z-th sub-problem, D z represents the demand of the z-th sub-problem, λ i represents the weight coefficient of the i-th type of resource, C zi represents the compatibility coefficient between the z-th sub-problem and the i-th type of resource, T z represents the scheduled completion time of the z-th sub-problem, ΔT zrepresents the deviation between the scheduled completion time and the actual completion time of the z-th sub-problem, where α and β are adjustment coefficients, and m i represents the total number of resource categories.

[0036] As a further solution of the present invention, using the sub-problem decomposition result, adjusting the constraint conditions of the sub-problems, and obtaining the Lagrange multiplier optimization result by iteratively solving the optimal solutions of the sub-problems and adjusting the Lagrange multipliers, the specific steps are as follows:

[0037] S501: Through the sub-problem decomposition result, in the cloud environment, adjust the constraint conditions of each sub-problem, analyze the adaptability of the constraint conditions in the real-time resource scheduling scenario, and generate a constraint adjustment record;

[0038] S502: Adopt the constraint adjustment record, capture the optimal solution of each sub-problem through an iterative solution method, adjust the solution of the sub-problem, match the dynamically changing resource demand and supply situation, and generate a sub-problem iterative solution set;

[0039] S503: Use the sub-problem iterative solution set to adjust the Lagrange multiplier to optimize the overall resource scheduling effect, verify the optimization of resource allocation, and generate a Lagrange multiplier optimization result.

[0040] As a further solution of the present invention, according to the Lagrange multiplier optimization result, perform dynamic allocation of resources for the predicted resource demand peaks and troughs, formulate and implement a resource allocation execution plan, and the specific steps to obtain a resource scheduling execution strategy are as follows:

[0041] S601: Through the Lagrange multiplier optimization result, analyze the predicted resource demand peak and trough periods, formulate a resource dynamic allocation strategy for the matching period, and set initial resource scheduling parameters to generate a resource dynamic allocation plan;

[0042] S602: Use the resource dynamic allocation plan to formulate an execution plan for resource allocation, including the schedule of resource scheduling and the priority of resource utilization, adjust and optimize the execution plan to check the resource utilization efficiency, and generate a resource allocation execution plan;

[0043] S603: Based on the resource allocation execution plan, implement resource scheduling, respond to market changes and real-time demands through real-time monitoring and adjustment strategies, verify the priority and efficiency of resource scheduling, and generate a resource scheduling execution strategy.

[0044] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0045] In the present invention, by deeply analyzing resource usage data, carefully classifying and statistically analyzing the usage frequency and duration of each resource, calculating the usage frequency distribution of the resources, it provides a precise decision-making support for resource scheduling. The classification and statistical method can effectively reveal the patterns and trends of resource usage, provide a scientific basis for formulating scheduling strategies, and achieve the maximization of resource utilization rate. Using the Hidden Markov Model to dynamically predict the resource usage status not only enhances the foresight and adaptability of resource scheduling, but also can adjust resource allocation pertinently according to the prediction results to cope with future demand changes. By using the Lagrangian relaxation technique, the complex resource scheduling problem is effectively decomposed into manageable sub-problems, which not only simplifies the problem processing flow, but also improves the solution efficiency and accuracy. The independent solution and optimization of each sub-problem make the overall resource scheduling more flexible and efficient, further improving the service quality and stability. This refined and intelligent resource scheduling strategy greatly enhances the dynamic management ability and response speed of resources, and provides users with more stable and efficient services. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a schematic diagram of the main steps of the present invention;

[0047] Figure 2 is a refined schematic diagram of S1 of the present invention;

[0048] Figure 3 is a refined schematic diagram of S2 of the present invention;

[0049] Figure 4 is a refined schematic diagram of S3 of the present invention;

[0050] Figure 5 is a refined schematic diagram of S4 of the present invention;

[0051] Figure 6 is a refined schematic diagram of S5 of the present invention;

[0052] Figure 7 is a refined schematic diagram of S6 of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0053] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0054] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more, unless otherwise specifically defined.

[0055] Please refer to Figure 1 , the present invention provides a technical solution, a resource scheduling method in a cloud environment, including the following steps:

[0056] S1: Based on the resource usage data in the cloud environment, classify and statistically analyze the usage frequency and duration of each resource, calculate the usage frequency distribution of differentiated resources, and obtain the resource usage frequency analysis result;

[0057] S2: According to the resource usage frequency analysis result, set the initial state probability matrix and state transition probability matrix of the hidden Markov model, and iteratively optimize the model parameters to obtain the hidden state model parameters;

[0058] S3: Use the hidden state model parameters to predict the state of resource usage in the next five years through the cloud environment, calculate the probability of state transition, and predict the state of resource demand according to the probability to obtain the resource demand prediction result;

[0059] S4: Through the resource demand prediction result, decompose the resource scheduling problem into multiple sub-problems by Lagrangian relaxation technology, set the resource type and demand for each sub-problem, and set the constraints and objective functions of the sub-problems to obtain the sub-problem decomposition result;

[0060] S5: Utilize the sub-problem decomposition result to adjust the constraint conditions of the sub-problems, iteratively solve the optimal solutions of the sub-problems, and adjust the Lagrange multipliers to obtain the Lagrange multiplier optimization result;

[0061] S6: According to the Lagrange multiplier optimization result, adjust the resource allocation strategy, dynamically allocate resources for the predicted resource demand peaks and troughs, formulate and implement the resource allocation execution plan to obtain the resource scheduling execution strategy.

[0062] The resource usage frequency analysis results include the average usage time, time period, and usage frequency distribution of each type of resource. The hidden state model parameters include the state transition probability matrix, emission probability matrix, and initial probability of the hidden state optimized according to the resource usage data. The resource demand prediction results include the probability prediction of the peak and trough of resource usage in the time period within the next five years and the corresponding resource demand quantities. The sub - problem decomposition results include the resource demand types, quantitative demand indicators, and corresponding constraint conditions of the sub - problems. The Lagrange multiplier optimization results include the adjusted constraint condition weights, number of iterations, and convergence status. The resource scheduling execution strategy includes calculating the resource allocation ratio, adjusting the network bandwidth priority, and dynamically allocating the storage space.

[0063] Please refer to Figure 2 , based on the resource usage data in the cloud environment, classify and count the usage frequency and duration of each resource, and calculate the usage frequency distribution of different resources. The specific steps to obtain the resource usage frequency analysis results are as follows:

[0064] S101: Extract the usage data of different resources from the cloud environment, record the access time point and end time point of each resource, and calculate the continuous usage duration of each resource. The execution process of generating the resource usage record is as follows;

[0065] Extract the usage data of different resources from the cloud environment. Extract the access records of each resource from the server log through API calls, including the unique identifier of the resource and the corresponding access time point and end time point. For each resource, calculate the continuous usage duration by subtracting the access time point from the end time point of each record to obtain the continuous duration of a single resource usage. The usage durations of all resources are accumulated to provide basic data for subsequent analysis. The usage record of each resource is then stored, formatted as resource identifier, access time point, end time point, and continuous usage duration, ensuring the integrity and accuracy of the data, laying a solid foundation for the next data screening and resource classification, and generating the resource usage record.

[0066] S102: Based on the resource usage record, screen and organize the data, group the resources according to the usage frequency and duration, and mark each group of data to distinguish different resource types. The execution process of generating the resource classification result is as follows;

[0067] Based on resource usage records, it involves screening and organizing the recorded data. The usage records of all resources are sorted according to the usage frequency, and the usage frequency is calculated by counting the number of accesses of each resource during the recording period. Combining with the continuous usage duration of the resources, the resources are further divided into multiple groups. The resources within each group not only have similar frequencies but also have a duration within a certain range. Each group of resources is marked to distinguish different resource types, such as computing resources, storage resources, etc. The classification criteria are formulated according to the characteristics and usage patterns of the resources to ensure the accuracy of each classification and generate the resource classification result.

[0068] S103: According to the resource classification result, the execution process of counting the usage frequency of each resource type, calculating the usage situation of each type of resource in different time periods, and organizing by category to generate the resource usage frequency analysis result is as follows;

[0069] According to the resource classification result, statistical analysis of the resource usage frequency is carried out. Each type of resource is analyzed one by one to count the total number of usage times during the recording period and the usage frequency in different time periods (such as weekdays, non - weekdays). Through comparative analysis, it can be found that the usage frequency of resources in specific time periods is higher than that in the remaining periods, which is of great significance for optimizing resource allocation and reducing costs. It will be presented in the form of a report, and the report includes the usage frequency and time distribution of various resources, facilitating more accurate decision - making by management and generating the resource usage frequency analysis result.

[0070] Please refer to Figure 3 , according to the resource usage frequency analysis result, setting the initial state probability matrix and state transition probability matrix of the hidden Markov model, and iteratively optimizing the model parameters to obtain the hidden state model parameters. The specific steps are as follows:

[0071] S201: Using the resource usage frequency analysis result and the entropy weight method, the execution process of assigning weights to the initial probabilities of the hidden states of the hidden Markov model to generate the initial state probability matrix is as follows;

[0072] Using the resource usage frequency analysis result as input data, the entropy weight method is applied to determine the initial probability distribution of each hidden state in the hidden Markov model. Calculate the entropy value of the usage frequency of each resource type to evaluate the amount of information and importance of each resource category in the total resource usage. The higher the entropy value, the more unstable the usage situation of the resource type and the richer the amount of information. According to the calculated entropy value, weights are assigned to each hidden state, and the weights reflect the influence degree of each resource category. After the weight calculation is completed, it is normalized to ensure that the sum of the initial probabilities of all states is 1, forming the initial state probability matrix.

[0073] The formula of the entropy weight method is as follows:

[0074]

[0075] Among them, P i is the initial state probability of each resource, and p ij represents the usage frequency of the i-th type of resource in the j-th time period. n j represents the total number of time periods, and m i represents the total number of resource categories. k represents the adjustment coefficient of information entropy, and e is the base of the natural logarithm.

[0076] Formula:

[0077]

[0078] It is used to calculate the initial state probability of each resource category. Considering the entropy weight of the usage frequency of each resource, the usage frequency p ij is determined, where i represents the resource category and j represents the time period. n j in the formula is the total number of time periods, and m i is the total number of resource categories. The adjustment coefficient k adjusts the weight according to the information entropy distribution, reflecting the influence size of the usage frequencies of different resources.

[0079] Specific value setting and calculation example: There are three resource categories set, and the usage frequencies p ij of each type of resource in four time periods are obtained through monitoring data. The following are the specific frequencies:

[0080] p 11 = 0.1, p 12 = 0.2, p 13 = 0.3, p 14 = 0.4;

[0081] p 21 = 0.3, p 22 = 0.3, p 23 = 0.2, p 24 = 0.2;

[0082] p 31 = 0.4, p 32 = 0.3, p 33 = 0.2, p 34 = 0.1;

[0083] Let k = 1. Based on the standard application of information entropy, the calculation of information entropy is as follows:

[0084]

[0085] Entropy1 = -(0.1log0.1 + 0.2log0.2 + 0.3log0.3 + 0.4log0.4);

[0086] Entropy2 = -(0.3 log 0.3 + 0.3 log 0.3 + 0.2 log 0.2 + 0.2 log 0.2);

[0087] Entropy3 = -(0.4 log 0.4 + 0.3 log 0.3 + 0.2 log 0.2 + 0.1 log 0.1);

[0088] Calculation result of information entropy:

[0089] Entropy1 ≈ 1.279

[0090] Entropy2 ≈ 1.211

[0091] Entropy3 ≈ 1.279

[0092] Calculate the initial state probability according to the entropy weight method:

[0093] e- E ntropy 1 ≈ e- 1.279 ≈ 0.278

[0094] e- E ntropy 2 ≈ e- 1.211 ≈ 0.298

[0095] e- E ntropy 3 ≈ e- 1.279 ≈ 0.278

[0096] Sum

[0097] Calculate the initial state probability P of each resource i :

[0098]

[0099] The calculation results show that the initial state probability of each resource category is directly related to the entropy of the usage frequency. The category with a lower entropy (more evenly distributed usage frequency) has a slightly higher initial probability. This reflects that the resources with more even usage are more important in the model, which can then guide the priority allocation of resources.

[0100] S202: Using the initial state probability matrix, using the resource usage frequency data, calculate the probability of resource state transition, set and adjust the state transition probability matrix, and optimize the transition probability. The execution process of generating the state transition probability matrix is as follows;

[0101] Based on the initial state probability matrix, resource usage frequency data is further used to calculate the transition probabilities between resource states. According to the statistical data of resource usage frequency, the frequencies of transitions between various resource states are determined, and the frequency data reflects the probability of a resource state changing from one to another. A preliminary state transition probability matrix is constructed based on the frequency data, and each element of the matrix represents the transition probability from one state to another. To optimize the transition probabilities, an iterative algorithm is used to adjust the matrix to ensure the accuracy and stability of model prediction. In each iteration, the parameters are adjusted according to the deviation between the model prediction results and the actual observation results. If there is a significant difference between the predicted transition frequency of a certain state and the actual observed frequency, the corresponding transition probability will be adjusted to reduce the deviation, and the state transition probability matrix is repeatedly optimized until it converges to a stable result, generating the state transition probability matrix.

[0102] S203: Based on the state transition probability matrix, the model parameters are optimized through iterative calculation, and the parameters of each iteration are adjusted to avoid prediction errors. The execution process of determining the optimal values of the model parameters and generating the hidden state model parameters is as follows;

[0103] Based on the state transition probability matrix, it involves optimizing the parameters of the hidden Markov model through iterative calculation. Initially, a set of model parameters is set, and the maximum likelihood estimation method is used to calculate the model parameters according to the existing data to maximize the probability of the observed data. In each iteration, the calculation results of the previous iteration are evaluated, and the parameters that cause prediction errors are identified and adjusted. The basis for adjustment includes the magnitude of the deviation, the statistical analysis results of the prediction errors, etc. The model performance is continuously monitored until the parameters converge to the optimal values. When the prediction error is the smallest, the model performs the best, generating the hidden state model parameters.

[0104] Please refer to Figure 4 , using the hidden state model parameters, the steps of predicting the state of resource usage in the next five years through the cloud environment, calculating the probability of state transition, and predicting the state of resource demand to obtain the resource demand prediction results are as follows:

[0105] S301: Through the hidden state model parameters, the prediction model of the resource usage state is initialized, and the model is configured to simulate the resource usage scenarios in the next five years. The execution process of generating the model prediction scenario is as follows;

[0106] S(t) = S(0)·e θ·t

[0107] Here, S(0) is the current resource usage state, and through the adjustment of the parameter θ, the state change in the next T years is simulated. The e θ·t in the formula represents the exponential change of the resource state over time, which is applicable to predicting the long-term trend of resource utilization.

[0108] To determine the value of θ, it is necessary to perform a least squares fitting based on historical data to calculate θ:

[0109]

[0110] where t i and S(t i ) are the historical time points and the corresponding resource usage status respectively, and are the means of the historical data.

[0111] Set a practical example. Assume that data for t i = [1, 2, 3, 4, 5] years is extracted from the historical data, and the corresponding resource status S(t i ) = [100, 120, 145, 175, 210]. The data can be used to calculate the specific value of θ and predict the resource usage status S(5) for the next five years.

[0112] S302: According to the model prediction scenario, use the hidden Markov model to calculate the state transition probability, predict the probability from the real-time state to the state in the next five years, and iteratively update the state probability. The execution process of generating the state transition probability prediction result is as follows;

[0113] According to the model prediction scenario, use the hidden Markov model to calculate the state transition probability. Define the current state of the model and the probabilities of each state transition. The states include different usage stages of resources. For each state, apply the hidden Markov model to calculate the probability of transitioning from the current state to each future state. The calculation is based on the state transition probability matrix. Adopt an iterative update method to consider new observation data to adjust and accurately predict the model. Each iteration optimizes the model parameters by comparing the changes in the state probabilities before and after to ensure the accuracy of the prediction and the adaptability of the model. The iterative process continues until the change in the state probability tends to be stable or reaches the preset number of iterations, generating the state transition probability prediction result. The formula used is:

[0114]

[0115] where, represents the transition probability of the future state, is the transition probability of the current state,

[0116] is the transition probability calculated based on the new observation data, and λ1, λ2, and λ3 are adjustment parameters.

[0117] S303: Based on the prediction results of the state transition probability, analyze and calculate the state changes of resource requirements, predict the usage requirements and supply situations of multiple resources, and collate the prediction information on resource requirements for the next five years. The execution process for generating the resource requirement prediction results is as follows;

[0118] Based on the prediction results of the state transition probability, further analyze and calculate the state changes of resource requirements. By evaluating the usage frequencies and demand trends of different resource types during the prediction period, predict the usage requirements and supply situations of multiple resources. During the process, factors such as seasonal variations of resources and market demand fluctuations will be considered, and complex variables will be integrated into the model to improve the accuracy and practicality of the prediction. According to the prediction results of the model, conduct a detailed prediction and analysis of the resource requirements for the next five years, which will include the expected resource supply situation, demand growth rate, and resource shortage risk, and have important reference value for resource planning and management, helping decision-makers formulate more effective resource allocation strategies, and generate the resource requirement prediction results. The formula used is:

[0119]

[0120] where, represents the predicted resource requirement, and are the maximum and minimum values of the resource requirement during the prediction period respectively, is the current resource requirement state, and α1, α2, and α3 are weight coefficients.

[0121] Please refer to Figure 5 , through the resource requirement prediction results, decompose the resource scheduling problem into multiple sub-problems by using the Lagrangian relaxation technique. The steps for setting the resource type, demand, and constraint conditions for the sub-problems to obtain the sub-problem decomposition results are as follows:

[0122] S401: Utilize the resource requirement prediction results, use the Lagrangian relaxation technique to decompose the resource scheduling problem into multiple independent sub-problems, determine the resource type and demand for each sub-problem, and set the initial constraint conditions. The execution process for generating the initial setting of the sub-problems is as follows;

[0123] Utilize the resource requirement prediction results, decompose the resource scheduling problem by using the Lagrangian relaxation technique. Decompose the resource scheduling problem into multiple independent sub-problems according to the prediction results. Each sub-problem corresponds to a specific resource type and demand quantity. Set the initial constraint conditions for each sub-problem. The constraints include the availability of the resource type, the limit of the demand quantity, and the constraint of the time window, which can simplify the complex resource scheduling problem and make it easier to manage and solve. After the initial setting is completed, the parameters and constraint conditions of each sub-problem are recorded to provide basic data for subsequent optimization and adjustment, and generate the initial setting of the sub-problems.

[0124] The formula of the Lagrangian relaxation technique is as follows:

[0125]

[0126] Among them, R z represents the resource allocation result of the z-th sub-problem, D z represents the demand of the z-th sub-problem, λ i represents the weight coefficient of the i-th type of resource, C zi represents the compatibility coefficient between the z-th sub-problem and the i-th type of resource, T z represents the scheduled completion time of the z-th sub-problem, ΔT z represents the deviation between the scheduled completion time and the actual completion time of the z-th sub-problem, α and β are adjustment coefficients respectively, and m i represents the total number of resource categories.

[0127]

[0128] The formula is used to calculate the resource allocation result R z of the z-th sub-problem, where D z represents the demand of the z-th sub-problem, which is obtained through data monitoring. After data monitoring, it is set that the demand of the z-th sub-problem is 200 units.

[0129] λ i represents the weight coefficient of the i-th type of resource, which is determined based on resource scarcity and cost analysis. If the first type of resource is relatively scarce, then λ1 = 0.5; if the second type of resource is relatively abundant, then λ2 = 0.2.

[0130] C zi represents the compatibility coefficient between the z-th sub-problem and the i-th type of resource, which is obtained through experimental data and usage efficiency evaluation. It is set that C z1 = 0.8 and C z2 = 0.3;

[0131] T z represents the scheduled completion time of the z-th sub-problem, which is set by the project management team. It is set that T z = 50 hours;

[0132] ΔT z represents the deviation between the scheduled completion time and the actual completion time of the z-th sub-problem, which is obtained through real-time monitoring of project tracking. It is set that ΔT z = -2 hours (completed ahead of schedule);

[0133] α and β are adjustment coefficients respectively, which are used to balance the influence of demand and time deviation. α = 0.1, β = 5, and the values are determined based on project data statistics and efficiency analysis.

[0134] Specific operation process:

[0135] Calculate the square root of the demand for the z-th sub-problem:

[0136] Calculate the weighted sum of resource compatibility:

[0137]

[0138] Calculate the ratio of demand to resource compatibility:

[0139]

[0140] Calculate the time ratio adjusted for time deviation:

[0141]

[0142] Apply the adjustment coefficient α:

[0143] 0.1 × 7.14 = 0.714

[0144] Resource allocation result R z :

[0145] R z = 30.74 + 0.714 = 31.454

[0146] The results show that according to the currently set parameters and conditions, the resource allocation result for the z-th sub-problem should be 31.454 units, reflecting the proportion of demand to resource compatibility and the optimized adjustment of the project completion time, supporting a more precise resource scheduling strategy.

[0147] S402: Adopt the sub-problem initialization setting, define the resource requirements and constraint information for each sub-problem, adjust and optimize the constraint conditions of the sub-problem, match the target situation of different resource types and requirements, and the execution process of generating the optimized result of the sub-problem constraints is as follows;

[0148] Adopt the sub-problem initialization setting, further define the resource requirements and constraint information for each sub-problem, adjust and optimize the initially set constraint conditions to match different resource types and specific demand targets. The process involves analyzing the matching degree between the resource supply situation and the demand prediction result, and adjusting the resource allocation strategy according to the actual situation. The purpose of optimization is to reduce resource waste and improve utilization efficiency, while ensuring that each sub-problem can meet specific requirements, generating the optimized result of each sub-problem constraint. The formula used is:

[0149]

[0150] Among them, represents the result of optimizing the sub-problem constraints, is the resource demand quantity, is the resource supply quantity, represents the difference between demand and supply, and δ1, δ2, and δ3 are weight parameters.

[0151] S403: According to the sub-problem constraint optimization result, perform resource scheduling and demand matching for each sub-problem, check the optimal allocation and utilization efficiency of resources, and obtain the resource allocation plan for the sub-problem. The execution process for generating the sub-problem decomposition result is as follows;

[0152] According to the sub-problem constraint optimization result, it involves performing resource scheduling and demand matching for each sub-problem. Through detailed resource analysis and scheduling strategies, check the optimal allocation and utilization efficiency of resources, evaluate whether the resource allocation plan for each sub-problem meets the expected efficiency and effect, optimize resource allocation, ensure the maximization of resource utilization, and considering the feasibility and cost-effectiveness of actual operations, the obtained resource allocation plan will optimize the use of resources on the premise of meeting the demand, generate the sub-problem decomposition result, and the formula used is:

[0153]

[0154] Among them, represents the resource allocation result, is the optimized resource demand, and are the actual and target utilization efficiencies respectively, is the difference between demand and allocation, and δ1, δ2, and δ3 are weight parameters.

[0155] Please refer to Figure 6 and use the sub-problem decomposition result to adjust the constraint conditions of the sub-problem. The steps for obtaining the Lagrangian multiplier optimization result by iteratively solving the optimal solution of the sub-problem and adjusting the Lagrangian multiplier are specifically as follows:

[0156] S501: Through the sub-problem decomposition result, in the cloud environment, adjust the constraint conditions of each sub-problem, analyze the adaptability of the constraint conditions in the real-time resource scheduling scenario, and the execution process for generating the constraint adjustment record is as follows;

[0157] Through the sub-problem decomposition result, adjust the constraint conditions of each sub-problem in the cloud environment, evaluate the adaptability of the constraint conditions of each sub-problem in the real-time resource scheduling scenario, dynamically adjust the constraint conditions by real-time monitoring of resource usage and demand changes to better adapt to the changes in resource supply and demand, analyze the impact of the constraint adjustment to determine the effectiveness and rationality of the adjustment, and generate the constraint adjustment record. The formula used is:

[0158]

[0159] Among them, represents the adjusted constraint value, is the current resource utilization rate, is the gap between resource demand and supply, ΔU g represents the change in resource utilization rate, e is the base of the natural logarithm, and α1, α2, and α3 are weight coefficients.

[0160] S502: Adopt constraint adjustment records, capture the optimal solution of each sub-problem through an iterative solution method, adjust the solution of the sub-problem, match the dynamically changing resource demand and supply situation, and the execution process of generating the iterative solution set of the sub-problem is as follows;

[0161] Adopt constraint adjustment records, capture the optimal solution of each sub-problem through an iterative solution method, define the resource demand and supply situation of each sub-problem, iteratively adjust the solution of each sub-problem to match the dynamically changing resource demand and supply situation. In each iteration, adjust the resource configuration and constraint conditions of the sub-problem, evaluate the adaptability and efficiency of the adjusted solution, and perform iterative solution until the optimal solution is found or the preset number of iterations is reached, generating the iterative solution set of the sub-problem. The formula used is:

[0162]

[0163] Among them, represents the optimized solution of the sub-problem, is the demand quantity, is the supply quantity, and represent the previous and current supply states respectively, and ρ1, ρ2, and ρ3 are adjustment coefficients.

[0164] S503: Utilize the iterative solution set of the sub-problem, adjust the Lagrange multiplier to optimize the overall resource scheduling effect, and verify the optimization of resource allocation. The execution process of generating the optimization result of the Lagrange multiplier is as follows;

[0165] Utilize the iterative solution set of the sub-problem, adjust the Lagrange multiplier to optimize the overall resource scheduling effect. By analyzing the resource allocation situation in the sub-problem solution set, evaluate whether the current Lagrange multiplier is suitable. According to the optimization requirements of resource allocation, adjust the value of the Lagrange multiplier. The adjustment aims to improve the efficiency and fairness of resource configuration. After each adjustment, verify the optimization of resource allocation to ensure that each adjustment is in the direction of improving resource utilization rate and meeting more demands, generating the optimization result of the Lagrange multiplier.

[0166] Please refer to Figure 7, according to the optimization results of Lagrange multipliers, for the predicted peak and trough of resource requirements, dynamically allocate resources, and formulate and implement a resource allocation execution plan. The specific steps to obtain the resource scheduling execution strategy are as follows:

[0167] S601: Through the optimization results of Lagrange multipliers, analyze the predicted peak and trough periods of resource requirements, formulate a dynamic resource allocation strategy for the matching periods, and set initial resource scheduling parameters. The execution process of generating a dynamic resource allocation plan is as follows;

[0168] Through the optimization results of Lagrange multipliers, analyze the predicted peak and trough periods of resource requirements. For these periods, formulate corresponding dynamic resource allocation strategies, including adjusting the resource supply volume to match the demand changes in different periods, and setting initial resource scheduling parameters such as the start time and end time of resource allocation, resource type, and allocation ratio, etc. These parameters will be used to construct a preliminary dynamic resource allocation plan. The plan is designed to maximize resource utilization and minimize costs. The initialized resource scheduling parameters and dynamic allocation strategies will be recorded to generate a dynamic resource allocation plan.

[0169] S602: Use the dynamic resource allocation plan to formulate an execution plan for resource allocation, including the schedule of resource scheduling and the priority of resource utilization. Adjust and optimize the execution plan to check the resource utilization efficiency. The execution process of generating a resource allocation execution plan is as follows;

[0170] Use the dynamic resource allocation plan to formulate an execution plan for resource allocation, including a detailed schedule of resource scheduling and the priority of resource utilization. The plan details the allocation time, type, quantity, and priority arrangement of resources. When formulating the execution plan, consider the availability of resources and the urgency of project requirements. Adjust and optimize the execution plan, and check the resource utilization efficiency to ensure that resource allocation can meet the efficiency and effectiveness in actual operations, and generate a resource allocation execution plan.

[0171] S603: Based on the resource allocation execution plan, implement resource scheduling, respond to market changes and real-time demands through real-time monitoring and adjustment strategies, and verify the priority and efficiency of resource scheduling. The execution process of generating a resource scheduling execution strategy is as follows;

[0172] Based on the resource allocation execution plan, implement resource scheduling, respond to market changes and real-time demands through real-time monitoring and adjustment strategies, including monitoring resource usage, market supply and demand dynamics, and project progress. Adjust the resource scheduling strategy according to the information to ensure the optimal configuration and utilization efficiency of resources. During the adjustment process, verify the priority and efficiency of resource scheduling to ensure that resource scheduling can quickly respond to changes in the market and project demands, and generate a resource scheduling execution strategy.

[0173] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A resource scheduling method in a cloud environment, characterized in that: The following steps are involved: Based on the resource usage data in the cloud environment, the usage frequency and duration of each resource are classified and counted, and the usage frequency distribution of differentiated resources is calculated to obtain the resource usage frequency analysis results; According to the resource usage frequency analysis results, the initial state probability matrix and the state transition probability matrix of the hidden Markov model are set, and the model parameters are iteratively optimized to obtain the hidden state model parameters; Using the hidden state model parameters, predicting the resource usage status in the next five years through the cloud environment, calculating the probability of state transition, predicting the resource demand status, and obtaining the resource demand prediction result; Based on the resource demand prediction results, the resource scheduling problem is decomposed into multiple sub-problems through Lagrangian relaxation technology, and resource types, demands and constraints are set for the sub-problems to obtain sub-problem decomposition results; Using the subproblem decomposition results, the constraint conditions of the subproblems are adjusted, the optimal solutions of the subproblems are solved iteratively, and the Lagrange multipliers are adjusted to obtain Lagrange multiplier optimization results; According to the Lagrange multiplier optimization result, resources are dynamically allocated according to the predicted resource demand peaks and valleys, a resource allocation execution plan is formulated and implemented, and a resource scheduling execution strategy is obtained.

2. The resource scheduling method in a cloud environment according to claim 1, characterized in that: The resource usage frequency analysis results include the average usage time, time period and usage frequency distribution of each type of resource. The hidden state model parameters include the state transition probability matrix, emission probability matrix and initial probability of the hidden state optimized according to the resource usage data. The resource demand prediction results include the probability prediction of the peak and trough of resource usage in the next five years and the corresponding resource demand. The sub-problem decomposition results include the resource demand type, quantitative demand index and corresponding constraints of the sub-problem. The Lagrange multiplier optimization results include the adjusted constraint weights, number of iterations and convergence status. The resource scheduling execution strategy includes computing resource allocation ratio, network bandwidth priority adjustment and dynamic allocation of storage space.

3. The resource scheduling method in a cloud environment according to claim 1, characterized in that: Based on the resource usage data in the cloud environment, the usage frequency and duration of each resource are classified and counted, and the usage frequency distribution of differentiated resources is calculated. The specific steps to obtain the resource usage frequency analysis results are as follows: Extract usage data of differentiated resources from the cloud environment, record the access time and end time of each resource, calculate the continuous usage duration of each resource, and generate resource usage records; Based on the resource usage records, the data is screened and sorted, the resources are grouped according to the usage frequency and duration, and each group of data is marked to distinguish the differentiated resource types, and a resource classification result is generated; According to the resource classification results, the usage frequency of each resource type is counted, the usage of each type of resource in a differentiated time period is calculated, and the usage is sorted by category to generate a resource usage frequency analysis result.

4. The resource scheduling method in a cloud environment according to claim 1, characterized in that: According to the resource usage frequency analysis results, the initial state probability matrix and state transition probability matrix of the hidden Markov model are set, and the model parameters are iteratively optimized to obtain the hidden state model parameters. Specifically, the steps are as follows: Using the resource usage frequency analysis result and the entropy weight method, weight allocation is performed on the hidden state initial probability of the hidden Markov model to generate an initial state probability matrix; Using the initial state probability matrix and resource usage frequency data, the probability of switching from a resource state is calculated, a state transition probability matrix is ​​set and adjusted, and the transition probability is optimized to generate a state transition probability matrix; Based on the state transition probability matrix, the model parameters are optimized through iterative calculation, and the parameters of each iteration are adjusted to avoid prediction errors, so as to determine the optimal values ​​of the model parameters and generate the hidden state model parameters.

5. The resource scheduling method in a cloud environment according to claim 4, characterized in that: The formula of the entropy weight method is as follows: Among them, P i is the initial state probability of each resource, p ij represents the usage frequency of the i-th resource in the j-th time period, n j Represents the total number of time periods, m i represents the total number of resource categories, k represents the adjustment coefficient of information entropy, and e is the base of the natural logarithm.

6. The resource scheduling method in a cloud environment according to claim 1, characterized in that: The steps of using the hidden state model parameters to predict the resource usage status in the next five years through the cloud environment, calculating the probability of state transition, predicting the resource demand status, and obtaining the resource demand prediction result are as follows: Initialize the prediction model of resource usage status through the hidden state model parameters, configure the model to simulate resource usage scenarios in the next five years, and generate model prediction scenarios; According to the model prediction scenario, the hidden Markov model is used to calculate the state transition probability, predict the probability from the real-time state to the state in the next five years, and iteratively update the state probability to generate the state transition probability prediction result; Based on the state transition probability prediction results, the state changes of resource demand are analyzed and calculated, the usage demand and supply of various resources are predicted, and the forecast information of resource demand in the next five years is sorted out to generate resource demand forecast results.

7. The resource scheduling method in a cloud environment according to claim 1, characterized in that: Based on the resource demand prediction results, the resource scheduling problem is decomposed into multiple sub-problems through Lagrangian relaxation technology, and resource types, requirements and constraints are set for the sub-problems. The specific steps for obtaining the sub-problem decomposition results are as follows: Using the resource demand prediction results and Lagrangian relaxation technology, the resource scheduling problem is decomposed into multiple independent sub-problems, the resource type and demand are determined for each sub-problem, and initial constraints are set to generate sub-problem initialization settings; Using the sub-problem initialization settings, defining resource requirements and constraint information for each sub-problem, adjusting and optimizing the constraint conditions of the sub-problem, matching the target conditions of differentiated resource types and requirements, and generating sub-problem constraint optimization results; According to the sub-problem constraint optimization results, resource scheduling and demand matching are performed for each sub-problem, the optimal allocation and utilization efficiency of resources are checked, and the resource allocation plan for the sub-problem is obtained to generate the sub-problem decomposition result.

8. The resource scheduling method in a cloud environment according to claim 7, characterized in that: The formula for the Lagrangian relaxation technique is as follows: Among them, R z represents the resource allocation result of the z-th sub-problem, D z represents the demand of the zth subproblem, λ i Represents the weight coefficient of the i-th resource, C zi represents the compatibility coefficient between the z-th sub-problem and the i-th resource, T z represents the scheduled completion time of the zth subproblem, ΔT z represents the deviation between the scheduled completion time and the real-time completion time of the z-th sub-problem, α and β are adjustment coefficients, respectively, and m i Represents the total number of resource categories.

9. The resource scheduling method in a cloud environment according to claim 1, characterized in that: The steps of adjusting the constraint conditions of the subproblems by using the subproblem decomposition results, solving the optimal solutions of the subproblems by iteration, and adjusting the Lagrange multipliers to obtain the Lagrange multiplier optimization results are as follows: Based on the sub-problem decomposition results, the constraint conditions of each sub-problem are adjusted through the cloud environment, the adaptability of the constraint conditions in the real-time resource scheduling scenario is analyzed, and a constraint adjustment record is generated; Adopting the constraint adjustment record, capturing the optimal solution of each sub-problem through an iterative solution method, adjusting the solution of the sub-problem, matching the dynamically changing resource demand and supply, and generating a sub-problem iterative solution set; The sub-problems are iteratively solved, the Lagrange multipliers are adjusted to optimize the overall resource scheduling effect, the optimization of resource allocation is verified, and the Lagrange multiplier optimization results are generated.

10. The resource scheduling method in a cloud environment according to claim 1, characterized in that: According to the Lagrange multiplier optimization results, resources are dynamically allocated according to the predicted resource demand peaks and valleys, and a resource allocation execution plan is formulated and implemented to obtain the resource scheduling execution strategy in the following steps: By using the Lagrange multiplier optimization results, the predicted resource demand peak and valley periods are analyzed, a resource dynamic allocation strategy for matching periods is formulated, and initialized resource scheduling parameters are set to generate a resource dynamic allocation plan; Utilize the dynamic resource allocation scheme to formulate a resource allocation execution plan, including a resource scheduling schedule and resource utilization priorities, adjust and optimize the execution plan to check resource utilization efficiency, and generate a resource allocation execution plan; Based on the resource allocation execution plan, resource scheduling is implemented, and market changes and real-time demands are responded to through real-time monitoring and adjustment of strategies, the priority and efficiency of resource scheduling are verified, and a resource scheduling execution strategy is generated.

Images