A cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment

Abstract

This invention discloses a cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment, comprising: Step 1: using the DBSCAN clustering algorithm to cluster cloud providers in the cloud environment into K clusters; Step 2: calculating the distance from the center of each of the K clusters to the cloud demander, and selecting the cloud provider in the cluster with the smallest distance as a candidate cloud provider; Step 3: selecting evaluation indicators from the non-functional attributes of the cloud providers, and assigning standardized weights to the evaluation indicators using the maximum range method to obtain the normalized final weight of each evaluation indicator; Step 4: ranking the candidate cloud providers using the TODIM decision-making method based on regret theory. This invention's method can evaluate and rank cloud providers according to the diverse resource needs and preferences of cloud demanders, and select the most suitable cloud provider to provide resources to them in the cloud platform.

Description

Technical Field

[0001] This invention belongs to the field of cloud computing technology and relates to a cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment. Background Technology

[0002] Since the concept of cloud computing was proposed in 2006, its development has gradually formed a three-tiered service model centered on data: Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS). The cloud computing market has been established and is developing rapidly. To meet the needs of cloud computing and adapt to future development, the concept of inter-cloud computing has been proposed. Inter-cloud computing is based on open collaboration among cloud service providers (CSPs). Through the deep integration of cloud resources from multiple parties, it facilitates developers to customize cloud services through "software-defined" methods, creating cloud value and realizing a new generation of cloud computing model characterized by "borderless services, inter-cloud collaboration, easy resource sharing, and exchangeable value." The inter-cloud environment provides a basic environment that supports fair transactions and interconnection between different CSPs. When cloud demanders send multi-granularity resource requests to the inter-cloud environment, the available resources of a single cloud provider in the inter-cloud environment can meet these requests. Therefore, the resource recommendation problem can be seen as the problem of selecting the most suitable cloud provider for cloud demanders. However, there are many cloud service providers with similar functions but different quality in the inter-cloud environment, making it a challenging task to find the best cloud service provider.

[0003] Common cloud provider selection algorithms include: machine learning-based algorithms, which select cloud providers by analyzing their historical service data, user behavior, and system performance; and collaborative filtering algorithms, which are mainly divided into two types: user-based collaborative filtering and content-item-based collaborative filtering. The core idea of ​​user-based collaborative filtering is to find other users with similar behaviors to the target user and then recommend cloud providers based on the behavior of these similar users. The main difference between content-item-based collaborative filtering and user-based collaborative filtering is that it calculates the similarity between content items, rather than the similarity between users, to obtain a recommendation list for a specific user.

[0004] The cloud provider recommendation methods mentioned above are mostly geared towards traditional cloud computing environments and do not consider the complexity of inter-cloud environments. They may face challenges in determining SLA parameters and lack comprehensiveness. Furthermore, the idea of ​​simply summing all SLA parameters and taking the average as the overall capability of the cloud provider has certain flaws. We need to consider that the relationships between SLA parameters may not be simple linear but complex nonlinear, and these variable interactions may affect the cloud provider's capability level. Small changes in the strength of factors can lead to significant differences in results; and the preferences of cloud demanders are not taken into account. Summary of the Invention

[0005] The purpose of this invention is to provide a cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment. This invention can evaluate and rank cloud providers according to the diverse resource needs and preferences of cloud demanders, and select the most suitable cloud provider to provide resources to them in the cloud platform.

[0006] This invention provides a cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment, including:

[0007] Step 1: Use the DBSCAN clustering algorithm to cluster cloud providers in the cloud environment to form K clusters;

[0008] Step 2: Calculate the distance from each of the K cluster centers to the cloud provider, and select the cloud provider in the cluster with the smallest distance as the candidate cloud provider;

[0009] Step 3: Select evaluation indicators from the non-functional attributes of the cloud provider, and assign standard weights to the evaluation indicators using the maximum range method to obtain the normalized final weight of each evaluation indicator.

[0010] Step 4: Rank the candidate cloud providers using the TODIM decision method based on regret theory.

[0011] Furthermore, step 1 specifically involves:

[0012] Step 1.1: Determine the cloud provider CSP in the cloud environment = {csp1, csp2, ... csp} n}, where each cloud provider is represented by a vector consisting of m non-functional attributes, and the CSP is a single instance of the cloud provider. i ={x i1 ,x i2 ,…x im}, i = 1…n;

[0013] Step 1.2: Input neighborhood parameters ε and MinPts, where ε is the neighborhood radius and MinPts is the threshold for neighborhood density;

[0014] Step 1.3: Initialize the core object collection Initialize the number of clusters K = 0, initialize the set of unvisited samples Γ = CSP, and then perform cluster partitioning.

[0015] Step 1.4: For all cloud provider CSPs i For i = 1…n, find the core object by following these steps;

[0016] The ε-neighbor subset N of sample cspi is calculated based on Euclidean distance. ε (csp i ), initially N ε (csp i Let ) be an empty set. Calculate the Euclidean distance D between sample cspi and other samples cspj according to the following formula. ij If CSP i and CSP j The distance is less than ε, i.e., D ij <ε, then csp j Add ε-neighborhood sample subset N ε (csp i )middle;

[0017]

[0018] Where, x im and x jm The sample CSPs are respectively i Other sample CSP j The numerical value composed of the m-th non-functional attribute;

[0019] Step 1.5: If the ε-neighborhood sample subset N ε (csp i The number of samples | N ε (csp i If the following formula is satisfied, then the sample csp will be... i Add to the core object sample set Ω:

[0020] |N ε (csp i )|≥MinPts (2) Step 1.6: If the core object set is empty, then end; otherwise, go to step 1.7.

[0021] Step 1.7: Randomly select a core object sample csp from the core object set Ω. i Initialize the category index K to the previous category index plus 1, and initialize the current cluster sample set C. k ={csp i}, find all from the core object csp iOther CSPs with achievable density j to all achievable density CSP j Added to the current cluster sample set, the current cluster generation is complete;

[0022] Step 1.8: Update the cluster sample set C k All cloud providers in the system are marked as accessed, and the core object CSP is set to accessed. i Removed from the core object collection Ω;

[0023] Step 1.9: Continue to randomly select other unvisited core objects in the core object set Ω, and repeat steps 1.7 and 1.8 until there are no unvisited core objects in the core object set Ω. The clustering algorithm ends, and a total of K clusters are formed.

[0024] Furthermore, step 2 specifically involves:

[0025] Step 2.1: Based on the clustering results in Step 1, calculate the center of each cluster: Center = {cet1,cet2,...,cet...} k}, k∈K,cet k cet is the center of the k-th cluster. k ={c k1 ,c k2 ,…c km}, c km This represents the value of the m-th non-functional attribute of the center of the k-th cluster;

[0026] Step 2.2: Represent the cloud demand provider using a vector as CSD = {d1, d2, ... d} m}, d m This represents the value of the m-th non-functional attribute of the cloud demand provider;

[0027] Step 2.3: Calculate the Euclidean distance from the cloud provider to each cluster center:

[0028]

[0029] Step 2.4: Select the cluster with the smallest distance from the cluster center as the candidate cluster, and the cloud providers in the candidate cluster will be selected as candidate cloud providers.

[0030] Furthermore, step 2.1 specifically calculates the CET of each cluster center according to the following formula. k The values ​​of m non-functional attributes:

[0031]

[0032] Where, x ymThis represents the value of the m-th non-functional attribute of the y-th cloud provider in the cluster, and z is the number of cloud providers in the k-th cluster.

[0033] Furthermore, step 3 specifically involves:

[0034] Step 3.1: Select q non-functional attributes from m non-functional attributes as evaluation indicators, and construct an indicator evaluation matrix based on the q non-functional attributes of p candidate cloud providers: V = [v ab ] p×q a = 1, 2, ..., p; b = 1, 2, ..., q; where v ab This represents the value of the b-th non-functional attribute of the a-th candidate cloud provider;

[0035] Step 3.2: Calculate the index weights using the Analytic Hierarchy Process (AHP). in The weight of the q-th evaluation index is calculated using the analytic hierarchy process.

[0036] Step 3.3: Calculate the index weights using the optimal and worst-case method. in The weight of the qth evaluation index is calculated using the optimal and inferior method.

[0037] Step 3.4: Calculate the index weights using the laboratory method. in The weight of the qth evaluation index obtained using the laboratory method;

[0038] Step 3.5: Calculate the final weight w for each evaluation index using the maximum deviation method. b The calculation formula is as follows:

[0039]

[0040] Where ξ is a Lagrange multiplier, a = 1, 2, ..., p;

[0041] Step 3.6: Take the partial derivative of equation (5) to obtain the following equation:

[0042]

[0043]

[0044] Simplifying equation (6) yields:

[0045]

[0046] Therefore, we get:

[0047]

[0048] Substituting equation (9) into equation (8), we get:

[0049]

[0050] make:

[0051]

[0052] get:

[0053]

[0054] Step 3.7: For w b The weights are then standardized to obtain the final weights.

[0055] Furthermore, step 3.7 applies the following formula to w. b Standardization process:

[0056]

[0057] Among them, w b * represents the final normalized weight of the b-th non-functional attribute.

[0058] Furthermore, step 4 specifically involves:

[0059] Step 4.1: Obtain the final attribute weights using the maximum deviation method, and calculate the relative weight w of each evaluation index relative to the reference index. br :

[0060]

[0061] w r =max{w b |b=1,2,...,q} (15)

[0062] Among them, w r The weight of the reference indicator;

[0063] Step 4.2: Calculate the cloud provider's CSP according to the following formula. a Compared to other cloud provider CSPs c Regarding evaluation index C b Advantage φ b (csp a ,csp c ):

[0064]

[0065] Where θ is the loss decay coefficient, set to 1;

[0066] Step 4.3: Calculate the cloud provider's CSP according to the following formula. a Compared to other cloud provider CSPs c Regarding the overall dominance φ(csp) of all evaluation indicators a ,csp c ):

[0067]

[0068] Step 4.4: Determine the cloud provider CSP a Compared to all other cloud providers, its overall advantage Φ(csp) a ), where Φ(csp) a ) + To achieve the maximum overall advantage, Φ(csp) a ) - The lowest overall advantage;

[0069]

[0070]

[0071]

[0072] Step 4.5: Calculate the euphoria value R(csp) for each cloud provider. a ), regret value G(csp) a ) and the joy-regret value Z(csp) a ):

[0073]

[0074]

[0075] Z(csp a ) = R(csp) a )+G(csp a ) (twenty three)

[0076] Step 4.6: Sort the cloud providers from highest to lowest based on their delight-regret scores and return the best cloud service provider.

[0077] This invention provides a cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment. This method evaluates and ranks cloud providers according to the diverse resource needs and preferences of cloud demanders, selecting the most suitable provider to offer resources within the cloud platform. By clustering cloud providers in the cloud environment and selecting candidate providers based on Euclidean distance, the number of cloud providers requiring screening is significantly reduced. Furthermore, the use of the maximum range method to determine criterion weights compensates for the shortcomings of single-value assignment and avoids the drawbacks of simple weighted summation in traditional combined weighting methods. Finally, integrating the risk-averse psychology of cloud demanders with classic decision-making methods makes this method more relevant to real-world scenarios. Compared with existing technologies, this invention can efficiently evaluate and recommend cloud providers in the cloud platform, ensuring the efficiency and real-time performance of the cloud platform. Attached Figure Description

[0078] Figure 1 A framework diagram for a cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment;

[0079] Figure 2 This is a flowchart of a cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment. Detailed Implementation

[0080] like Figure 1 The diagram shown is a framework diagram of a cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment according to the present invention, which includes three roles: cloud demander, cloud provider, and cloud platform.

[0081] Cloud providers are primarily responsible for publishing information on the resources they can provide, including computing, storage, networking, and data, as well as information on the quality of their historical services on the cloud platform.

[0082] Cloud demand providers: primarily responsible for publishing demand information and preference information for various aspects of resources, including computing, storage, network, and data, on the cloud platform;

[0083] Cloud Platform: Collects, processes, and stores supply and demand information published by cloud providers and cloud demanders. It recommends cloud providers based on a multi-criteria decision-making approach, considering the supply and demand relationship between cloud providers and demanders.

[0084] The interaction between the three roles is as follows: cloud providers and cloud demanders publish supply and demand information to the cloud platform. The cloud platform receives the information and evaluates and ranks cloud providers according to the supply and demand relationship, and returns the results to cloud providers and cloud demanders. Then, the cloud provider provides the requested resources to the cloud demander.

[0085] like Figure 2 As shown, the present invention provides a cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment, comprising:

[0086] Step 1: Use the DBSCAN clustering algorithm to cluster cloud providers in the cloud environment to form K clusters. Step 1 specifically involves:

[0087] Step 1.1: Determine the cloud provider CSP in the cloud environment = {csp1, csp2, ..., csp} n}, where each cloud provider is represented by a vector consisting of m non-functional attributes, and the CSP is a single instance of the cloud provider. i ={x i1 ,x i2 ,…x im}, i = 1…n;

[0088] In specific implementation, the non-functional attributes include: response time, availability, throughput, reliability, success rate, latency, etc. This invention selects m non-functional attributes.

[0089] Step 1.2: Input neighborhood parameters ε and MinPts, where ε is the neighborhood radius and MinPts is the threshold for neighborhood density;

[0090] Step 1.3: Initialize the core object collection Initialize the number of clusters K = 0, initialize the set of unvisited samples Γ = CSP, and then perform cluster partitioning.

[0091] Step 1.4: For all cloud provider CSPs i For i = 1…n, find the core object by following these steps;

[0092] The ε-neighbor subset N of sample cspi is calculated based on Euclidean distance. ε (csp i ), initially N ε (csp i Let ) be an empty set. Calculate the Euclidean distance D between sample cspi and other samples cspj according to the following formula. ij If CSP i and CSP j The distance is less than ε, i.e., D ij <ε, then csp j Add ε-neighborhood sample subset N ε (csp i )middle;

[0093]

[0094] Where, x im and x jm The sample CSPs are respectively i Other sample CSP jThe numerical value composed of the m-th non-functional attribute;

[0095] Step 1.5: If the ε-neighborhood sample subset N ε (csp i The number of samples | N ε (csp i If the following formula is satisfied, then the sample csp will be... i Add to the core object sample set Ω:

[0096] |N ε (csp i )|≥MinPts (2)

[0097] Step 1.6: If the core object collection is empty, then end; otherwise, go to step 1.7.

[0098] Step 1.7: Randomly select a core object sample csp from the core object set Ω. i Initialize the category index K to the previous category index plus 1, and initialize the current cluster sample set C. k ={csp i}, find all from the core object csp i Other CSPs with achievable density j to all achievable density CSP j Added to the current cluster sample set, the current cluster generation is complete;

[0099] Step 1.8: Update the cluster sample set C k All cloud providers in the system are marked as accessed, and the core object CSP is set to accessed. i Removed from the core object collection Ω;

[0100] Step 1.9: Continue to randomly select other unvisited core objects in the core object set Ω, and repeat steps 1.7 and 1.8 until there are no unvisited core objects in the core object set Ω. The clustering algorithm ends, and a total of K clusters are formed.

[0101] Step 2: Calculate the distance from each of the K cluster centers to the cloud provider, and select the cloud provider in the cluster with the smallest distance as the candidate cloud provider. Step 2 specifically involves:

[0102] Step 2.1: Based on the clustering results in Step 1, calculate the center of each cluster: Center = {cet1,cet2,...,cet...} k}, k∈K,cet k cet is the center of the k-th cluster. k ={c k1 ,c k2 ,…ckm}, c km This represents the value of the m-th non-functional attribute of the center of the k-th cluster;

[0103] The CET of each cluster center is calculated according to the following formula. k The values ​​of m non-functional attributes:

[0104]

[0105] Where, x ym This represents the value of the m-th non-functional attribute of the y-th cloud provider in the cluster, and z is the number of cloud providers in the k-th cluster.

[0106] Step 2.2: Represent the cloud demand provider using a vector as CSD = {d1, d2, ... d} m}, d m This represents the value of the m-th non-functional attribute of the cloud demand provider;

[0107] Step 2.3: Calculate the Euclidean distance from the cloud provider to each cluster center:

[0108]

[0109] Step 2.4: Select the cluster with the smallest distance from the cluster center as the candidate cluster, and the cloud providers in the candidate cluster will be selected as candidate cloud providers.

[0110] After step 2, you only need to select a cloud provider from the cluster with the smallest distance to provide resources to the cloud demander. Assuming there are p cloud providers in the cluster, this reduces the number of cloud providers that need to be selected and saves a lot of time.

[0111] Step 3: Select evaluation indicators from the non-functional attributes of the cloud provider, and assign standardized weights to the evaluation indicators using the maximum range method to obtain the normalized final weight of each evaluation indicator. Step 3 specifically involves:

[0112] Step 3.1: Select q non-functional attributes from m non-functional attributes as evaluation indicators, and construct an indicator evaluation matrix based on the q non-functional attributes of p candidate cloud providers: V = [v ab ] p×q a = 1, 2, ..., p; b = 1, 2, ..., q; where v ab This represents the value of the b-th non-functional attribute of the a-th candidate cloud provider;

[0113] Step 3.2: Calculate the index weights using the Analytic Hierarchy Process (AHP). in The weight of the q-th evaluation index is calculated using the analytic hierarchy process.

[0114] Step 3.3: Calculate the index weights using the optimal and worst-case method. in The weight of the qth evaluation index is calculated using the optimal and inferior method.

[0115] Step 3.4: Calculate the index weights using the laboratory method. in The weight of the qth evaluation index obtained using the laboratory method;

[0116] Step 3.5: Calculate the final weight w for each evaluation index using the maximum deviation method. b The calculation formula is as follows:

[0117]

[0118] Where ξ is a Lagrange multiplier, a = 1, 2, ..., p;

[0119] Step 3.6: Take the partial derivative of equation (5) to obtain the following equation:

[0120]

[0121]

[0122] Simplifying equation (6) yields:

[0123]

[0124] Therefore, we get:

[0125]

[0126] Substituting equation (9) into equation (8), we get:

[0127]

[0128] make:

[0129]

[0130] get:

[0131]

[0132] Step 3.7: For w b The weights are then standardized to obtain the final weights.

[0133] In specific implementation, w is calculated according to the following formula. b Standardization process:

[0134]

[0135] Among them, w b * represents the final normalized weight of the b-th non-functional attribute.

[0136] Step 4: Rank the candidate cloud providers using the TODIM decision-making method based on regret theory. Step 4 specifically involves:

[0137] Step 4.1: Obtain the final attribute weights using the maximum deviation method, and calculate the relative weight w of each evaluation index relative to the reference index. br :

[0138]

[0139] w r =max{w b |b=1,2,...,q} (15)

[0140] Among them, w r The weight of the reference indicator;

[0141] Step 4.2: Calculate the cloud provider's CSP according to the following formula. a Compared to other cloud provider CSPs c Regarding evaluation index C b Advantage φ b (csp a ,csp c ):

[0142]

[0143] Where θ is the loss decay coefficient, set to 1;

[0144] Step 4.3: Calculate the cloud provider's CSP according to the following formula. a Compared to other cloud provider CSPs c Regarding the overall dominance φ(csp) of all evaluation indicators a ,csp c ):

[0145]

[0146] Step 4.4: Determine the cloud provider CSP a Compared to all other cloud providers, its overall advantage Φ(csp) a ), where Φ(csp) a ) + To achieve the maximum overall advantage, Φ(csp) a ) - The lowest overall advantage;

[0147]

[0148]

[0149]

[0150] Step 4.5: Calculate the euphoria value R(csp) for each cloud provider. a ), regret value G(csp) a ) and the joy-regret value Z(csp) a ):

[0151]

[0152]

[0153] Z(csp a ) = R(csp) a )+G(csp a ) (twenty three)

[0154] Step 4.6: Sort the cloud providers from highest to lowest based on their delight-regret scores and return the best cloud service provider.

[0155] The above description is only a preferred embodiment of the present invention and is not intended to limit the ideas of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment, characterized in that, include: Step 1: Use the DBSCAN clustering algorithm to cluster cloud providers in the cloud environment to form K clusters; Step 2: Calculate the distance from each of the K cluster centers to the cloud provider, and select the cloud provider in the cluster with the smallest distance as the candidate cloud provider; Step 3: Select evaluation indicators from the non-functional attributes of the cloud provider, and assign standard weights to the evaluation indicators using the maximum range method to obtain the normalized final weight of each evaluation indicator. An evaluation matrix is ​​constructed based on q non-functional attributes of p candidate cloud providers: ;in This represents the value of the b-th non-functional attribute of the a-th candidate cloud provider; Step 4: Rank the candidate cloud providers using the TODIM decision-making method based on regret theory. Step 4 specifically involves: Step 4.1: Obtain the final attribute weights using the maximum deviation method, and calculate the relative weight w of each evaluation index relative to the reference index. br : (14) (15) Among them, w r The weight of the reference indicator; Step 4.2: Calculate the cloud provider's CSP according to the following formula. a Compared to other cloud provider CSPs c Regarding evaluation index C b Advantages : (16) in, The loss degradation coefficient is set to 1; Step 4.3: Calculate the cloud provider's CSP according to the following formula. a Compared to other cloud provider CSPs c Regarding the overall dominance of all evaluation indicators : (17) Step 4.4: Determine the cloud provider CSP a Compared to all other cloud providers, its overall advantages ,in To maximize overall advantages, The lowest overall advantage; (18) (19) (20) Step 4.5: Calculate the euphoria value R(csp) for each cloud provider. a ), regret value G(csp) a ) and the joy-regret value Z(csp) a ): (21) (22) (23) Step 4.6: Sort the cloud providers from highest to lowest based on their joy-regret scores and return the best cloud service provider.

2. The cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment as described in claim 1, characterized in that, Step 1 is as follows: Step 1.1: Determine the cloud provider in the cloud environment Each cloud provider is represented by a vector consisting of m non-functional attributes. ; Step 1.2: Input neighborhood parameters ε and MinPts, where ε is the neighborhood radius and MinPts is the threshold for neighborhood density; Step 1.3: Initialize the core object collection Initialize the number of clusters K=0, and initialize the set of unvisited samples. Cluster partitioning ; Step 1.4: For all cloud providers Find the core object by following these steps; Calculate sample CSP based on Euclidean distance. i ε-neighborhood sample subset Initially For an empty set, calculate the sample CSP according to the following formula. i Other sample CSP j Euclidean distance D ij If CSP i and CSP j The distance is less than ε, that is Then csp j Add ε-neighborhood sample subset middle; （1） in, and The sample CSPs are respectively i Other sample CSP j The numerical value composed of the m-th non-functional attribute; Step 1.5: If the ε-neighborhood sample subset Number of samples If the following formula is satisfied, then the sample csp will be... i Add to core object sample collection middle: （2） Step 1.6: If the core object collection is empty, then end; otherwise, go to step 1.

7. Step 1.7: Randomly select a core object sample csp from the core object set Ω. i Initialize the category index K to the previous category index plus 1, and initialize the current cluster sample set C. k ={csp i }, find all from the core object csp i Other CSPs with achievable density j to all achievable density CSP j Added to the current cluster sample set, the current cluster generation is complete; Step 1.8: Update the cluster sample set C k All cloud providers in the system are marked as accessed, and the core object CSP is set to accessed. i Removed from the core object collection Ω; Step 1.9: Continue to randomly select the core object set. For other unvisited core objects, repeat steps 1.7 and 1.8 until there are no unvisited core objects in the core object set Ω. The clustering algorithm ends, and a total of K clusters are formed.

3. The cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment as described in claim 1, characterized in that, Step 2 is as follows: Step 2.1: Based on the clustering results from Step 1, calculate the center of each cluster. ,cet k As the center of the k-th cluster, , This represents the value of the m-th non-functional attribute of the center of the k-th cluster; Step 2.2: Represent cloud demand providers using vector representation. , This represents the value of the m-th non-functional attribute of the cloud demand provider; Step 2.3: Calculate the Euclidean distance from the cloud provider to each cluster center: （3） Step 2.4: Select the cluster with the smallest distance from the cluster center as the candidate cluster, and the cloud providers in the candidate cluster will be selected as candidate cloud providers.

4. The cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment as described in claim 3, characterized in that, Step 2.1 Specifically, calculate the CET of each cluster center according to the following formula. k The values ​​of m non-functional attributes: （4） in, This represents the value of the m-th non-functional attribute of the y-th cloud provider in the cluster, and z is the number of cloud providers in the k-th cluster.

5. The cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment as described in claim 1, characterized in that, Step 3 specifically involves: Step 3.1: Select q non-functional attributes from m non-functional attributes as evaluation indicators, and construct an indicator evaluation matrix based on the q non-functional attributes of p candidate cloud providers: ;in This represents the value of the b-th non-functional attribute of the a-th candidate cloud provider; Step 3.2: Calculate the index weights using the Analytic Hierarchy Process (AHP). ,in The weight of the q-th evaluation index is calculated using the analytic hierarchy process. Step 3.3: Calculate the index weights using the optimal and worst-case method. ,in The weight of the qth evaluation index is calculated using the optimal and inferior method. Step 3.4: Calculate the index weights using the laboratory method. ,in The weight of the qth evaluation index obtained using the laboratory method; Step 3.5: Calculate the final weight of each evaluation indicator using the maximum deviation method. The calculation formula is as follows: （5） in, For Lagrange multipliers, ; Step 3.6: Take the partial derivative of equation (5) to obtain the following equation: （6） （7） Simplifying equation (6) yields: （8） Therefore, we get: （9） Substituting equation (9) into equation (8), we get: （10） make: （11） get: （12） Step 3.7: For The weights are then standardized to obtain the final weights.

6. The cross-domain resource recommendation method based on multi-criteria decision-making in a cloud environment as described in claim 5, characterized in that, Step 3.7 Apply the following formula to Standardization process: (13) in, The final weight for the normalization of the b-th non-functional attribute.

Images