Multi-stage matching method based on preference in cloud environment

Abstract

The preference-based multi-stage matching method in the cloud environment comprises the following steps: screening out cloud service providers meeting hard resource requirements of cloud service consumers through a rule matching method, and storing the cloud service providers in a candidate set S0; the performance of the cloud service providers in the candidate set S0 on the aspect of soft requirements is evaluated, the cloud service providers which do not meet the requirements are excluded, the set S0 is updated, and an improved MICe method and a UMAP dimension reduction technology are adopted to process correlation evaluation of high-dimensional data; determining a Pareto optimal solution set through a multi-objective optimization technology, and further identifying a cloud service provider set S1 which most meets the requirement in the candidate set S0 by comparing the dominating relationship of solutions; s2, calculating the matching degree of each cloud service provider in the S1 by using a weighted summation method or a fuzzy comprehensive evaluation method according to the personalized weight setting of the cloud service consumers, and ranking according to the matching degree so as to determine a final service matching result.

Description

Technical Field

[0001] The present invention relates to the technical field of cloud computing, and in particular to a preference-based multi-stage matching method in a cloud environment. Background Art

[0002] In today's digital age, cloud computing has undoubtedly become a key force driving the rapid development of information technology. It has a series of unique advantages, such as no need for high initial investment, relatively low operating costs, and high scalability, which has triggered a far-reaching change in the field of information technology. However, the rapid evolution of cloud computing has brought many opportunities, but it has also inevitably spawned a series of severe challenges. With the continuous acceleration of its development, the demand for cloud services has shown an explosive growth trend, which is undoubtedly a huge pressure for individual cloud service providers (CSPs).

[0003] During representative special events such as Taobao's "Double Eleven" and Walmart's "Black Friday", a large number of users simultaneously flood into the platform, and the demand for cloud service resources instantly reaches a peak. At this time, the resource reserves and supply capabilities of a single CSP are often stretched to the limit, making it difficult to meet such a large and concentrated resource demand in a short period of time, which in turn leads to the inability to provide users with the required key resources in a timely manner, seriously affecting user experience and normal business operations. At the same time, users are often trapped in the dilemma of "platform lock-in" when using cloud services. Once an enterprise chooses a specific cloud service provider, as the business is carried out on the platform, the enterprise's dependence on the platform will continue to deepen, gradually forming a situation of over-reliance on a single provider. This situation not only limits the flexibility of enterprises in choosing cloud services, but also may make it difficult for enterprises to respond quickly and effectively when faced with problems such as declining service quality and unreasonable price increases, thereby hindering the healthy development of the cloud market.

[0004] In order to effectively resolve these thorny problems, academia and industry have turned their attention to the in-depth exploration of the Cloud 2.0 architecture. In this process, a variety of cloud cooperation models have emerged, among which InterCloud, SuperCloud and CloudService Broker have attracted much attention. The core goal of these models is to build a new sharing model, break the resource limitations of a single CSP, promote extensive and in-depth collaboration between cloud service providers, and thus achieve mutual benefit and win-win results for all parties. Among these models, JointCloud has emerged with its unique architectural design and innovative ideas. It innovatively introduces the two key concepts of "vertical integration" and "horizontal cooperation". "Vertical integration" focuses on breaking through the communication barriers between heterogeneous cloud stack layers to achieve seamless interaction and collaborative work between layers; "horizontal cooperation" focuses on promoting close cooperation between different CSPs and integrating the resource advantages of all parties.

[0005] However, despite the many impressive advantages and broad application prospects of JointCloud, it still inevitably faces a series of complex challenges in its actual operating environment. With the widespread application of JointCloud, it has attracted more and more consumers to join in, which has directly led to an exponential growth in the types and complexity of consumer needs. From the perspective of consumers, the preferences of different consumers vary significantly. Some consumers regard the reputation of the provider as the primary consideration when choosing cloud services. They believe that a good reputation is an important guarantee for service quality and stability; while other consumers pay more attention to the specific attributes of cloud services, such as data processing capabilities, security levels, and customization. These attributes are directly related to whether their business can be carried out efficiently and smoothly. From the perspective of providers, as the market scale continues to expand, the types of resources available are becoming more and more numerous. How to accurately screen out services that perfectly match the diverse needs of consumers in this complex ocean of resources has become a very challenging task.

[0006] In the field of service matching, although many existing methods have attempted to solve the matching problem between consumers and service providers to a certain extent, they all have limitations to varying degrees. The method based on multi-criteria decision making (MCDM) attempts to comprehensively consider multiple criteria such as cost, performance, and security to rank service providers. However, in practical applications, it faces the difficulty of accurately determining the weights of each criterion and is unable to deal with subjective user preferences. The matching method based on trust and reputation mainly selects cloud services by evaluating the reliability of providers. However, this method often focuses too much on the reputation of providers and largely ignores the personalized preferences and specific resource requirements of users. This may result in the selected service being reputable but unable to meet the actual business needs of users. The matching method based on quality of service (QoS) focuses on parameters such as availability, response time, and cost, and uses optimization techniques to meet user-defined QoS constraints. However, in actual operations, it is difficult to fully capture the subtle differences in user preferences for different resources and data relevance, and cannot achieve truly accurate matching. Although the preference-based matching mechanism attempts to improve user satisfaction to a certain extent, it often ignores the specific resource requirements of users and lacks effective response strategies in dealing with whether users provide preference information, resulting in unsatisfactory matching results. The matching method based on utility maximization evaluates the overall utility by assigning weights to each criterion, but it also ignores the key factor of user preference, which may lead to a large deviation between the evaluation results and the actual expectations of users. Although the matching method based on uncertainty uses fuzzy technology and reference point method to deal with the uncertainty in cloud service selection, it also fails to fully consider user preferences in the processing process and is prone to overfitting historical preference data, which means that when the market environment or user needs change, this method may not be able to timely and accurately identify those new services that are more beneficial to users.

[0007] In the JointCloud environment, there is an urgent need to design a new and more effective matching method that can accurately grasp users' diverse preferences while fully considering resource requirements, thereby achieving more efficient and accurate matching between consumers and cloud service providers, and promoting the sustainable and healthy development of the cloud service market under the JointCloud architecture. Summary of the invention

[0008] In order to solve the above technical problems, the purpose of the present invention is to provide a preference-based multi-stage matching method in a cloud environment.

[0009] The present invention provides a preference-based multi-stage matching method in a cloud environment, comprising:

[0010] Step 1: Filter out cloud service providers that meet the hard resource requirements of cloud service consumers through rule matching methods and save them in the candidate set S0;

[0011] Step 2: Evaluate the performance of cloud service providers in the candidate set S0 on soft requirements, exclude cloud service providers that do not meet the requirements, and update the set S0. The soft requirements include cost-effectiveness and the relevance of data resources. Use the improved MICe method and UMAP dimensionality reduction technology to process the relevance evaluation of high-dimensional data.

[0012] Step 3: Determine the Pareto optimal solution set through multi-objective optimization technology, and further identify the cloud service provider set S1 that best meets the requirements in the candidate set S0 by comparing the dominance relationship of the solutions;

[0013] Step 4: According to the personalized weight setting of the cloud service consumer, use the weighted sum method or the fuzzy comprehensive evaluation method to calculate the matching degree of each cloud service provider in S1, and rank them accordingly to determine the final service matching result.

[0014] The present invention discloses a preference-based multi-stage matching method in a cloud environment. In view of the fact that some service matching methods mainly focus on the inherent attributes of the service itself and often ignore the problem of user preferences, a multi-stage matching strategy is used to integrate user preferences, provider reputation and data relevance to achieve more accurate matching. Rule-based matching is used to narrow the search space, a reverse cost algorithm is used for scoring, and the MICe indicator is used to evaluate data similarity. The Pareto front is determined by a dominance-based method, and a weighting technique customized for user preferences is used. Compared with existing technologies, the method of the present invention can integrate user preferences, service provider reputation and data resource relevance to ensure more accurate and efficient matching in a cloud environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The present invention is a flowchart of a preference-based multi-stage matching method in a cloud environment. DETAILED DESCRIPTION

[0016] The present invention proposes a preference-based multi-stage matching method in a cloud environment, which aims to improve matching efficiency, quickly exclude unqualified cloud service providers, achieve personalized matching, comprehensively consider user preferences, service provider reputation and data resource relevance, reduce computational complexity, use maximum information coefficient estimator (MICe) and dimension reduction technology to improve processing speed, enhance dynamic adaptability, flexibly respond to changes in consumer demand, strengthen privacy protection, implement reverse cost method to avoid users from disclosing sensitive information, and provide comprehensive comprehensive evaluation, using coverage decision method and entropy weight fuzzy comprehensive evaluation to comprehensively evaluate the performance of candidate cloud service providers. Through these goals, the present invention aims to provide an efficient, accurate, personalized and user-friendly cloud service matching solution to improve cloud service selection quality and user satisfaction while ensuring the security of user privacy.

[0017] The present invention discloses a preference-based multi-stage matching method in a cloud environment, which involves the following three roles: a cloud service provider, a cloud service consumer, and a cloud platform.

[0018] Cloud service provider: mainly responsible for publishing available resource information including computing, storage, network, data, etc. in the cloud platform and information on the quality of service related to the historical services provided;

[0019] Cloud service consumers: select and use the most appropriate cloud services on the cloud platform based on their personal needs and preferences;

[0020] Yunji Platform: Collect, process and store the supply and demand information released by cloud service providers and cloud service consumers. Recommend cloud service providers based on the supply and demand relationship between cloud service providers and cloud service consumers.

[0021] The interaction between the three roles is as follows: cloud service providers and cloud service consumers publish supply and demand information to the cloud platform respectively. The cloud platform receives the information and evaluates and ranks the cloud service providers based on the supply and demand relationship, and returns the results to the cloud service providers and cloud service consumers. Next, the cloud service provider provides the requested resources to the cloud service consumer.

[0022] like Figure 1 As shown, a preference-based multi-stage matching method in a cloud environment of the present invention includes:

[0023] Step 1: Filter out cloud service providers that meet the hard resource requirements of cloud service consumers through rule matching methods and save them in the candidate set S0. Specifically:

[0024] Step 1.1: Define the hard resource requirements of the cloud service consumer as:

[0025] U=(U compute ,U network,U storage )

[0026] Among them, U compute , U network and U storage Represent the demand for computing, network and storage resources respectively; the resource provision vector of the cloud service provider is represented as P a =(P a,compute ,P a,network ,P a,storage ), where P a,compute represents the computing resources of the a-th cloud service provider, P a,nework represents the network resources of the a-th cloud service provider, P a,storage Represents the storage resource amount of the a-th cloud service provider.

[0027] Step 1.2: Define rules based on the needs of cloud service consumers:

[0028] R=(R compute ,R network ,R storage )

[0029] Among them, R compute , R network and R storage Represents the matching rules for computing, network, and storage resources respectively; the matching rules are defined as:

[0030]

[0031] in, Respectively represent the minimum and maximum acceptable computing resources, Respectively represent the minimum and maximum acceptable network resources, and Respectively represent the minimum and maximum acceptable storage resources.

[0032] Step 1.3: Collect the needs of cloud service consumers and the resource provision capabilities of cloud service providers as input, and check whether the resources of each cloud service provider comply with the rules in the order of submission. If any resource does not comply with the rules, adopt an early termination strategy and immediately check the next cloud service provider; if all resources comply with the rules, add the cloud service provider to the set S0.

[0033] Step 1.4: If the set is empty, output a message that no cloud service provider meets the requirements; otherwise, return the set S0, which contains cloud service providers that meet the hard resource requirements.

[0034] Step 2: Evaluate the performance of cloud service providers in the candidate set S0 on soft requirements, exclude cloud service providers that do not meet the requirements, and update the set S0. The soft requirements include cost-effectiveness and the relevance of data resources. Use the improved MICe method and UMAP dimensionality reduction technology to process the relevance evaluation of high-dimensional data to reduce the computational complexity. Specifically:

[0035] Step 2.1: When cloud service consumers choose cloud service providers, revenue is a key consideration. The revenue formula for cloud service consumers is defined as:

[0036] Pro=Pri-Cost

[0037] Among them, Pro represents revenue, Pri represents sales price, and Cost represents cost. In practice, CSCs usually pursue higher revenue, but not all cloud service consumers CSCs are willing to disclose the total revenue of their projects. Since the sales price of cloud service consumers is fixed and only known to themselves, the reverse cost scoring method is used to define the profit score. For the a-th cloud service provider, its cost is Cost a , then the corresponding reverse cost score, that is, the profit score is: This means that the lower the costs, the higher the reverse cost score and the higher the potential profit score.

[0038] Step 2.2: Compare the benefits of different cloud service providers and normalize the cost scores to a range of 0 to 1 according to the following formula:

[0039]

[0040] Among them, C' a represents the normalized cost score of the a-th cloud service provider, C a represents the cost score of the a-th cloud service provider, C min and C max Represent the smallest and largest cost scores among all cloud service providers, respectively.

[0041] Through this formula, the reverse cost score of each cloud service provider CSP is mapped to a unified scale. Finally, the normalized reverse cost score C' a It is used as the final revenue score Proi, so that the revenue brought by different CSPs to CSC can be evaluated and compared under the same criteria;

[0042] Step 2.3: Evaluate the relevance between the data provided by the cloud service provider and the cloud service consumer task, and record the cloud service consumer task data as The data provided by the cloud service provider is denoted as Each data point and These vectors are all high-dimensional vectors that can fully describe the data characteristics of a specific cloud business. For example, they may contain information in multiple dimensions, such as the size of the data, the type distribution of the data, and the frequency of data access.

[0043] Step 2.4: Choose cosine similarity As a metric to measure the similarity between data points, however, for the convenience of subsequent calculations, it is converted into cosine distance, the formula is:

[0044]

[0045] Cosine distance It can more intuitively reflect the degree of difference between data points. The larger the value, the greater the difference between the data points; the smaller the value, the more similar the data points are.

[0046] Step 2.5: Efficiently search for the nearest neighbors of data points in high-dimensional space, and add each data point to the Annoy index. Annoy approximates the distribution of data points by generating multiple random projection trees. Each projection tree divides and organizes the data points from a different perspective, so that similar data points can be quickly located when searching for the nearest neighbors, thereby greatly improving the search efficiency.

[0047] Step 2.6: For each data point Find its k nearest neighbors in the Annoy index and calculate the local scale parameter ρ according to the following formula i :

[0048]

[0049] Among them, ρ i Represents data points The distance to its nearest neighbor; Represents data points The k nearest neighbors of ρ i It can reflect the density and distribution of data points in a local range, and play a key adjustment role in the subsequent similarity calculation, ensuring that the similarity calculation can adapt to the local characteristics of different data points.

[0050] Step 2.7: Use the calculated local scale parameter ρ i and cosine distance distance (x i ,x j ) to calculate the data points and Smoothing distance d ij , the formula is:

[0051]

[0052] The calculation of smoothed distance can adjust the original cosine distance to make the distance metric smoother and more continuous, reduce distance mutations caused by factors such as uneven distribution of data points or noise, and thus more accurately reflect the true relationship between data points.

[0053] Step 2.8: By smoothing the distance d ij Perform exponential transformation to calculate the similarity p ij , the formula is:

[0054]

[0055] Among them, σ i is a smoothing parameter, which is obtained by solving the following formula:

[0056]

[0057] Similarity p ij The larger the value, the more data points and The higher the similarity between them, the more effective it is to quantify the local neighborhood relationship between data points in high-dimensional space.

[0058] Step 2.9: Combine the similarities between each data point and other data points to construct a high-dimensional neighborhood graph. This high-dimensional neighborhood graph can intuitively display the similarity structure between data points and provide a basis for subsequent analysis and processing.

[0059] Step 2.10: In order to reduce the computational complexity while retaining the neighborhood structure relationship between data points, it is necessary to perform dimensionality reduction on the high-dimensional data, randomly initialize the data point Y in the low-dimensional space, and then calculate the data point in the low-dimensional space according to the following formula: and The similarity between:

[0060]

[0061] Use gradient descent to minimize the cross entropy loss function between the high-dimensional and low-dimensional representations:

[0062]

[0063] Through continuous iterative optimization, the low-dimensional representation can retain the neighborhood structure information of the high-dimensional data as much as possible, thereby reducing the difficulty and computational cost of data processing without losing too much information. The data in the low-dimensional space obtained after the dimensionality reduction of the cloud service consumer task data in the high-dimensional space is represented as Y A, the data in the low-dimensional space obtained after the data dimensionality reduction provided by the cloud service provider is represented as Y B .

[0064] Step 2.11: MICe is a method for evaluating two random variables Y A and Y B The MICe value is obtained by calculating the sample feature matrix M(D) and the maximum mutual information value; for random variables based on (Y A ,Y B ), define the sample feature matrix M(D), whose elements are expressed as follows:

[0065]

[0066] Among them, I * (D,m,l) is the maximum mutual information value on the m*l grid, where m and l represent the number of rows and columns of the grid respectively; the MICe value is obtained by summing the elements of the sample feature matrix, that is:

[0067] MICe(D)=max m,l M(D) m,l

[0068] The higher the MICe value, the higher the A and Y B The stronger the average mutual information between them, the stronger the correlation between them.

[0069] Step 2.12: Since MICe calculation involves calculating mutual information on a fixed grid size, the computational complexity is high. In order to improve computational efficiency, Bayesian optimization is used to search for the optimal grid configuration, and the objective function is defined as:

[0070] f(m,l)=-MICe(Y A ; Y B )

[0071] Because Bayesian optimization is usually a minimization problem, the opposite of the MICe value is taken as the objective function. The goal is to find the grid configuration that maximizes the MICe value by minimizing this objective function.

[0072] Set the search space, select the grid parameter m*l as the optimization variable, and determine the scope and granularity of the search.

[0073] Firstly, the Gaussian process is used to construct the prior distribution and calculate the initial MICe value. That is, under the initial grid partitioning parameters, the MICe value on the cloud service consumer and cloud service provider data sets is calculated.

[0074] When observing the data After that, the prior distribution is updated to the posterior distribution, taking into account the existence of Gaussian noise Compute the parameters of the joint and posterior distributions.

[0075] Finally, the expected improvement strategy is used to select the next sampling point and calculate the expected improvement value:

[0076]

[0077] Among them, f(m,l + ) is the current optimal solution. The expected value is calculated through the posterior distribution, and the input point that maximizes the expected improvement value is selected as the new sampling point.

[0078] Step 2.13: Repeat step 2.12 to continuously update the grid parameters and calculate the MICe value until the convergence condition is met, that is, the change in the objective function value is less than the set threshold ∈. At this time, the grid configuration obtained is the optimal configuration, thereby improving the efficiency and accuracy of the MICe calculation.

[0079] Step 2.14: Based on the cloud service consumers' requirements for benefit scores and data relevance, the set S0 obtained in step 1 is further screened, cloud service providers that do not meet the requirements are excluded, and the set S0 is updated.

[0080] Step 3: Determine the Pareto optimal solution set, namely the Pareto front, through multi-objective optimization technology, and further identify the cloud service provider set S1 that best meets the requirements in the candidate set S0 by comparing the dominance relationship of the solutions. Specifically:

[0081] Step 3.1: In order to comprehensively evaluate cloud service providers in a cloud environment, a multi-objective function is defined:

[0082] minf(X)=(-f1(X),-f2(X),…,-f6(X))

[0083] here is a decision vector variable, where X arepresents the decision vector of the ath cloud service provider, f1(X) represents the amount of computing resources provided by the cloud service provider, which is a key indicator to measure whether the CSP can meet the computing task requirements of the CSC; f2(X) represents the amount of network resources, which is related to the speed and stability of data transmission; f3(X) is the amount of storage resources, which determines the amount of data that the cloud service provider can store; f4(X) is the profit score, which reflects the economic value of the cloud service provider to the cloud service consumer; f5(X) reflects the data relevance, which means the degree of relevance between the data provided by the cloud service provider and the tasks of the cloud service consumer; f6(X) is the reputation score, which represents the reputation and word of mouth of the cloud service provider in the market. Through such a comprehensive objective function, the performance of CSP in multiple dimensions can be comprehensively evaluated in order to select the CSP that best suits the CSC.

[0084] The candidate set S0 consists of the remaining CSPs from the first two stages (the first stage is based on rule matching screening and the second stage is soft demand calculation). The first stage has eliminated those CSPs that do not meet the hard resource requirements through rule matching, and the second stage calculates the soft demand information in the remaining CSPs. After these two stages of screening, the remaining CSPs meet some of the requirements of the CSC to a certain extent, but there may still be some redundancy or CSPs that have room for improvement in multi-objective optimization. These remaining CSPs together constitute the candidate set of this stage, waiting for further screening and optimization.

[0085] Step 3.2: Each solution X in the candidate set S0 a , each solution X of the current Pareto front P needs to be p Compare to determine the dominant relationship between them; if for all objective functions, X a Not inferior to X in some respects p , and is better than X in at least one objective function p , then we call it X a DominateX p , denoted by X a <X p ; Assume X a Not dominated, marked as dominated = false; traverse each X in the Pareto front P p If X is found p DominateX a , then immediately mark dominated as true and stop p comparison, because at this time it has been determined that X a is dominated; on the contrary, if X a DominateX p , then X pAdd to the to-be-removed list to_remove, because the dominated solutions will be removed in the subsequent update of the Pareto front.

[0086] Step 3.3: After completing a solution X in the candidate set S0 a After judging the dominance relationship with all solutions in the Pareto front P, if X a If it is not dominated, then it is necessary to update the Pareto frontier P and traverse each solution X in the to-be-removed column to_remove. r , remove them from P, because these solutions have been replaced by the new solution X a Controlled by X a Added to P, P includes all the undominated solutions so far, forming a new Pareto frontier; by continuously updating the Pareto frontier, the final matching set can include the solutions of the best cloud service providers in the sense of multi-objective optimization. That is, these solutions have achieved a relatively optimal balance between various objective functions, and no other solution can improve the value of a certain objective function without reducing the value of other objective functions.

[0087] Step 3.4: After traversing all solutions of the candidate set S0, the Pareto frontier P obtained at this time is the final matching set; this matching set includes the cloud service provider set S1 that is most suitable for cloud service consumers in the multi-objective optimization framework according to the defined objective function and the judgment of the dominant relationship. This Pareto frontier P is output to provide a basis for subsequent decision-making. Cloud service consumers can choose the cloud service provider that best meets their needs and preferences based on this matching set to cooperate, thereby obtaining better experience and benefits in cloud computing services.

[0088] Step 4: According to the personalized weight setting of the cloud service consumer, use the weighted sum method or the fuzzy comprehensive evaluation method to calculate the matching degree of each cloud service provider in S1, and rank them accordingly to determine the final service matching result, specifically:

[0089] Step 4.1: In the cloud computing environment, the dimensions of various evaluation indicators such as computing, network and storage resources are diverse and the data distribution is uneven, but the interval is relatively fixed. In order to make these indicators comparable, the upper and lower limits of the resource indicators of each cloud service provider are clearly set according to the specific needs of cloud service consumers. Resource indicators include computing resources, network resources and storage resources. For example, for computing resources, if the CSC requires the processing capacity to be within a certain range, the upper and lower limits of the computing resource indicators can be determined accordingly. The resource indicators are processed using the Min-Max standardization method. When the maximum and minimum values ​​of the indicator are different, the resource indicator value is mapped to the interval of 0 to 1. If the maximum and minimum values ​​are equal, the indicator value is set to 1. In this way, indicator data with different ranges and distributions can be converted into a unified standard, which is convenient for subsequent calculation and comparison. It is already standardized data. It only needs to standardize the reputation score of the cloud service provider to ensure that all indicators used to evaluate the cloud service provider are under the same metric.

[0090] Step 4.2: When cloud service consumers clearly express the importance of each evaluation indicator, the optimized expert scoring method is used to invite experts in related fields to score each indicator. These experts evaluate the performance of cloud service providers on each indicator based on their professional knowledge and experience. Experts may score based on the stability of computing resources, network resource transmission speed, and storage resource security of cloud service providers. Then, the weight of each evaluation indicator is determined through normalization processing. The formula is:

[0091]

[0092] Among them, x cd is the standardized value of the dth evaluation indicator of the cth cloud service provider, ω cd It represents the weight of the dth evaluation indicator of the cth cloud service provider, and N is the total number of evaluation indicators. The sum of the weights of each evaluation indicator of each service provider is 1, and the weight distribution can accurately reflect the preferences of cloud service consumers. The evaluation indicators include computing resources, network resources, storage resources, profit score, data relevance and reputation score.

[0093] Calculate the final score based on the weights:

[0094]

[0095] H d represents the final score of the d-th cloud service provider; X cd It refers to the normalized cth evaluation index of the dth cloud service provider. All cloud service providers are ranked according to the final score.

[0096] Step 4.3: If the cloud service consumer does not disclose the importance of each indicator, in order to objectively determine the weight, the entropy weight method is used to weight each evaluation indicator, and a normalized matrix Z is constructed based on the previously standardized data, where Z cd The element represents the standardized data of the cth cloud service provider on the dth indicator. The information entropy e of each indicator is calculated according to the definition of entropy. d :

[0097]

[0098] Among them, M is the number of cloud service providers, and information entropy reflects the discreteness of indicator data. The larger the information entropy, the smaller the amount of information provided by the indicator, and the smaller the weight should be; conversely, the smaller the information entropy, the greater the weight.

[0099] The weights of each evaluation index are calculated by information entropy according to the following formula:

[0100]

[0101] Step 4.4: Define the membership function using the fuzzy comprehensive evaluation method. Since the data distribution of each indicator is uncertain and fuzzy, it is necessary to calculate relevant statistics, such as mean, variance, etc., to understand the central tendency and dispersion of the data; according to these statistical characteristics, fit different types of membership functions, such as triangular membership function, trapezoidal membership function, etc. By comparing the degree of fit between each membership function and the data distribution, the function with the highest fit is selected as the final membership function; the indicator values ​​of the cloud service provider can be converted into fuzzy membership values, and a comprehensive evaluation matrix R is constructed, where the element r cd represents the fuzzy membership value of the c-th cloud service provider on the d-th indicator.

[0102] Step 4.5: Calculate the fuzzy comprehensive evaluation value of each cloud service provider through fuzzy weighted summation. The formula is:

[0103]

[0104] S c represents the fuzzy comprehensive evaluation value of the cth cloud service provider, w d is the weight of the dth average indicator. The fuzzy membership value of each indicator of each cloud service provider and the corresponding weight are weighted and summed to obtain an evaluation value that comprehensively reflects the overall performance of the cloud service provider.

[0105] Step 4.6: Based on the calculated fuzzy comprehensive evaluation value, all cloud service providers in S1 are sorted from high to low. The sorted cloud service provider list can clearly show the order of advantages and disadvantages of each cloud service provider in meeting the needs of cloud service consumers, so that cloud service consumers can make the best choice according to their own needs and choose the cloud service provider that best meets their expectations for cooperation.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A preference-based multi-stage matching method in a cloud environment, characterized in that: include: Step 1: Filter out cloud service providers that meet the hard resource requirements of cloud service consumers through rule matching methods and save them in the candidate set S0; Step 2: Evaluate the performance of cloud service providers in the candidate set S0 on soft requirements, exclude cloud service providers that do not meet the requirements, and update the set S0. Soft requirements include cost-effectiveness and relevance of data resources. The improved MICe method and UMAP dimension reduction technique are used to handle the correlation evaluation of high-dimensional data; Step 3: Determine the Pareto optimal solution set through multi-objective optimization technology, and further identify the cloud service provider set S1 that best meets the requirements in the candidate set S0 by comparing the dominance relationship of the solutions; Step 4: According to the personalized weight setting of the cloud service consumer, use the weighted sum method or the fuzzy comprehensive evaluation method to calculate the matching degree of each cloud service provider in S1, and rank them accordingly to determine the final service matching result.

2. The multi-stage matching method based on preference in a cloud environment according to claim 1 is characterized in that: The step 1 is specifically as follows: Step 1.1: Define the hard resource requirements of the cloud service consumer as: U=(U compute ,IN network ,IN storage ) Among them, U compute , U network and U storage Represent the demand for computing, network and storage resources respectively; the resource provision vector of the cloud service provider is represented as P a =(P a,compute ,P a,network ,P a,storage ), where P a,compute represents the computing resources of the a-th cloud service provider, P a,network represents the network resources of the a-th cloud service provider, P a,storage represents the storage resource amount of the a-th cloud service provider; Step 1.2: Define rules based on the needs of cloud service consumers: R=(R compute ,R network ,R storage ) Among them, R compute , R network and R storage Represents the matching rules for computing, network, and storage resources respectively; the matching rules are defined as: in, Respectively represent the minimum and maximum acceptable computing resources, Respectively represent the minimum and maximum acceptable network resources, and Respectively represent the minimum and maximum acceptable storage resources; Step 1.3: Collect the cloud service consumer's requirements and the cloud service provider's resource provision capabilities as input, and check whether the resources of each cloud service provider meet the rules in the order of submission. If any resource does not meet the rules, adopt the early termination strategy and immediately check the next cloud service provider; if all resources meet the rules, add the cloud service provider to the set S; Step 1.4: If the set is empty, output a message that no cloud service provider meets the requirements; otherwise, return the set S, which contains cloud service providers that meet the hard resource requirements.

3. The preference-based multi-stage matching method in a cloud environment according to claim 1, characterized in that: The step 2 is specifically as follows: Step 2.1: When cloud service consumers choose cloud service providers, revenue is a key consideration. The revenue formula for cloud service consumers is defined as: Pro=Pri-Cost Among them, Pro represents revenue, Pri represents sales price, and Cost represents cost; Since the selling price of cloud service consumers is fixed and known only to themselves, the reverse cost scoring method is used to define the profit score. For the a-th cloud service provider, its cost is Cost a , then the corresponding reverse cost score, that is, the profit score is: This means that the lower the costs, the higher the reverse cost score, and the higher the potential profit score; Step 2.2: Compare the benefits of different cloud service providers and normalize the cost scores to a range of 0 to 1 according to the following formula: Among them, C' a represents the normalized cost score of the a-th cloud service provider, C a represents the cost score of the a-th cloud service provider, C min and C max denote the smallest and largest cost scores among all cloud service providers, respectively; Step 2.3: Evaluate the relevance between the data provided by the cloud service provider and the cloud service consumer task, and record the cloud service consumer task data as The data provided by the cloud service provider is denoted as Each data point and They are all high-dimensional vectors that can fully describe the data characteristics of specific cloud services; Step 2.4: Choose cosine similarity As a metric to measure the similarity between data points, however, for the convenience of subsequent calculations, it is converted into cosine distance, the formula is: Cosine distance It can more intuitively reflect the degree of difference between data points. The larger the value, the greater the difference between the data points; the smaller the value, the more similar the data points are. Step 2.5: Efficiently search for the nearest neighbors of data points in high-dimensional space and add each data point to the Annoy index. Annoy approximates the distribution of data points by generating multiple random projection trees. Step 2.6: For each data point Find its k nearest neighbors in the Annoy index and calculate the local scale parameter ρ according to the following formula i : Among them, ρ i Represents data points The distance to its nearest neighbor; Represents data points The k nearest neighbors of ; Step 2.7: Use the calculated local scale parameter ρ i and cosine distance cos ine distance (x i ,x j ) to calculate the data points and Smoothing distance d ij , the formula is: Step 2.8: By smoothing the distance d ij Perform exponential transformation to calculate the similarity p ij , the formula is: Among them, σ i is a smoothing parameter, which is obtained by solving the following formula: Step 2.9: Combine the similarities between each data point and other data points to construct a high-dimensional neighborhood graph; Step 2.10: In order to reduce the computational complexity while retaining the neighborhood structure relationship between data points, it is necessary to perform dimensionality reduction on the high-dimensional data, randomly initialize the data point Y in the low-dimensional space, and then calculate the data point in the low-dimensional space according to the following formula: and The similarity between: Use gradient descent to minimize the cross entropy loss function between the high-dimensional and low-dimensional representations: Through continuous iterative optimization, the low-dimensional representation can retain the neighborhood structure information of the high-dimensional data as much as possible, thereby reducing the difficulty and computational cost of data processing without losing too much information. The data in the low-dimensional space obtained after the dimensionality reduction of the cloud service consumer task data in the high-dimensional space is represented as Y A , the data in the low-dimensional space obtained after the data dimensionality reduction provided by the cloud service provider is represented as Y B ; Step 2.11: MICe is a method for evaluating two random variables Y A and Y B The MICe value is obtained by calculating the sample feature matrix M(D) and the maximum mutual information value; for random variables based on (Y A ,Y B ), define the sample feature matrix M(D), whose elements are expressed as follows: Among them, I * (D,m,l) is the maximum mutual information value on the m*l grid, where m and l represent the number of rows and columns of the grid respectively; the MICe value is obtained by summing the elements of the sample feature matrix, that is: MICe(D)=max m,l M(D) m,l The higher the MICe value, the higher the A and Y B The stronger the average mutual information between them, the stronger the correlation between them; Step 2.12: Use Bayesian optimization to search for the optimal grid configuration and define the objective function as: f(m,l)=-MICe(Y A ;Y B ) Set the search space, select the grid parameter m*l as the optimization variable, and determine the search range and granularity; First, the Gaussian process is used to construct the prior distribution and calculate the initial MICe value. That is, the MICe value on the cloud service consumer and cloud service provider data sets is calculated under the initial grid partitioning parameters. When observing the data After that, the prior distribution is updated to the posterior distribution, taking into account the existence of Gaussian noise Compute parameters of the joint and posterior distributions; Finally, the expected improvement strategy is used to select the next sampling point and calculate the expected improvement value: Among them, f(m,l + ) is the current optimal solution. The expected value is calculated through the posterior distribution, and the input point that maximizes the expected improvement value is selected as the new sampling point; Step 2.13: Repeat step 2.12 to continuously update the grid parameters and calculate the MICe value until the convergence condition is met, that is, the change in the objective function value is less than the set threshold ∈. At this time, the grid configuration obtained is the optimal configuration, thereby improving the efficiency and accuracy of the MICe calculation; Step 2.14: Based on the cloud service consumers' requirements for benefit scores and data relevance, the set S obtained in step 1 is further screened, cloud service providers that do not meet the requirements are excluded, and the set S is updated.

4. The preference-based multi-stage matching method in a cloud environment according to claim 1, characterized in that: The step 3 is specifically as follows: Step 3.1: In order to comprehensively evaluate cloud service providers in a cloud environment, a multi-objective function is defined: minf(X)=(-f1(X),-f2(X),…,-f6(X)) here is a decision vector variable, where X a represents the decision vector of the a-th cloud service provider, f1(X) represents the amount of computing resources provided by the cloud service provider; f2(X) represents the amount of network resources, which is related to the speed and stability of data transmission; f3(X) is the amount of storage resources, which determines the amount of data that the cloud service provider can store; f4(X) is the profit score, which reflects the economic value of the cloud service provider to the cloud service consumer; f5(X) reflects the data relevance, which means the degree of relevance between the data provided by the cloud service provider and the tasks of the cloud service consumer; f6(X) is the reputation score, which represents the credibility and reputation of the cloud service provider in the market; Step 3.2: Each solution X in the candidate set S0 a , each solution X of the current Pareto front P needs to be p Compare to determine the dominant relationship between them; if for all objective functions, X a Not inferior to X in some respects p , and is better than X in at least one objective function p , then we call it X a DominateX p , denoted by X a <X p ; Assume X a Not dominated, marked as dominated = false; traverse each X in the Pareto front P p If X is found p DominateX a , then immediately mark dominated as true and stop p On the contrary, if X a DominateX p , then X p Add to the to-be-removed list to_remove, because the dominated solution will be removed when the Pareto front is subsequently updated; Step 3.3: After completing a solution X in the candidate set S0 a After judging the dominance relationship with all solutions in the Pareto front P, if X a If it is not dominated, then it is necessary to update the Pareto frontier P and traverse each solution X in the to-be-removed column to_remove. r , remove it from P; X a Added to P, P includes all the undominated solutions so far, forming a new Pareto frontier; by continuously updating the Pareto frontier, the final matching set can include the solutions of the best cloud service providers in the sense of multi-objective optimization; Step 3.4: After traversing all solutions of the candidate set S0, the Pareto frontier P obtained at this time is the final matching set; this matching set includes the cloud service provider set S1 that is most suitable for cloud service consumers in the multi-objective optimization framework according to the defined objective function and the judgment of the dominant relationship. This Pareto frontier P is output to provide a basis for subsequent decision-making. Cloud service consumers can choose the cloud service provider that best meets their needs and preferences based on this matching set to cooperate, thereby obtaining better experience and benefits in cloud computing services.

5. The preference-based multi-stage matching method in a cloud environment according to claim 4, characterized in that: The step 4 is specifically as follows: Step 4.1: According to the specific needs of cloud service consumers, clearly set the upper and lower limits of each cloud service provider's resource indicators. Resource indicators include computing resources, network resources, and storage resources. Use the Min-Max normalization method to process resource indicators. When the maximum and minimum values ​​of the indicator are different, map the resource indicator value to the range of 0 to 1. If the maximum and minimum values ​​are equal, set the indicator value to 1. The data is already standardized. We only need to standardize the reputation scores of cloud service providers to ensure that all indicators used to evaluate cloud service providers are under the same metric. Step 4.2: When cloud service consumers clearly express the importance of each evaluation indicator, the optimized expert scoring method is used to invite experts in related fields to score each indicator. These experts evaluate the performance of cloud service providers on each indicator based on their professional knowledge and experience. Experts may score based on the stability of computing resources, network resource transmission speed, and storage resource security of cloud service providers. Then, the weight of each evaluation indicator is determined through normalization processing. The formula is: Among them, x cd is the standardized value of the dth evaluation indicator of the cth cloud service provider, ω cd Then represents the weight of the dth evaluation indicator of the cth cloud service provider, and N is the total number of evaluation indicators. The sum of the weights of each evaluation indicator of each service provider is 1, and the weight distribution can accurately reflect the preferences of cloud service consumers. The evaluation indicators include computing resources, network resources, storage resources, profit score, data relevance, and reputation score. Calculate the final score based on the weights: H d represents the final score of the d-th cloud service provider; X cd It refers to the normalized cth evaluation index of the dth cloud service provider. All cloud service providers are ranked according to the final score. Step 4.3: If the cloud service consumer does not disclose the importance of each indicator, in order to objectively determine the weight, the entropy weight method is used to weight each evaluation indicator, and a normalized matrix Z is constructed based on the previously standardized data, where Z cd The element represents the standardized data of the cth cloud service provider on the dth indicator. The information entropy e of each indicator is calculated according to the definition of entropy. d : Where M is the number of cloud service providers, and information entropy reflects the discreteness of the indicator data. The larger the information entropy, the smaller the amount of information provided by the indicator, and the smaller the weight should be; conversely, the smaller the information entropy, the greater the weight; The weights of each evaluation index are calculated by information entropy according to the following formula: Step 4.4: Define the membership function using the fuzzy comprehensive evaluation method. Since the data distribution of each indicator is uncertain and fuzzy, it is necessary to calculate the relevant statistics first to understand the central tendency and dispersion of the data. According to these statistical characteristics, fit different types of membership functions. By comparing the degree of fit between each membership function and the data distribution, select the function with the highest fit as the final membership function. The indicator values ​​of the cloud service provider can be converted into fuzzy membership values, and a comprehensive evaluation matrix R is constructed, where the element r cd represents the fuzzy membership value of the c-th cloud service provider on the d-th indicator; Step 4.5: Calculate the fuzzy comprehensive evaluation value of each cloud service provider through fuzzy weighted summation. The formula is: S c represents the fuzzy comprehensive evaluation value of the cth cloud service provider, w d is the weight of the dth average indicator. The fuzzy membership value of each indicator of each cloud service provider and the corresponding weight are weighted and summed to obtain an evaluation value that comprehensively reflects the overall performance of the cloud service provider. Step 4.6: Based on the calculated fuzzy comprehensive evaluation value, all cloud service providers in S1 are sorted from high to low. The sorted cloud service provider list can clearly show the order of advantages and disadvantages of each cloud service provider in meeting the needs of cloud service consumers, so that cloud service consumers can make the best choice according to their own needs and choose the cloud service provider that best meets their expectations for cooperation.

Images