Service combination optimization method under incomplete information

Abstract

The invention discloses a service combinatorial optimization method under incomplete information, and belongs to the technical field of calculation application of combinatorial optimization models.According to the method, a non-dominated solution set obtained in the combinatorial optimization problem of the incomplete information is converted into a solution for solving a 0-1 integer programming problem, and the method comprises the steps of initialization, generation of random feasible indexes, calculation and calculation of the non-dominated solution set. And generating a random weight vector, solving a 0-1 integer programming problem, and solving an approximate optimal combination by applying a decision criterion. The method provides an effective technical means for solving combinatorial optimization under incomplete information, solves the problems of low speed, low approximation algorithm accuracy and the like of combinatorial optimization under incomplete information solving at present, can be widely applied to the field of engineering, and has wide application prospects. And particularly, the problems of boxing, cloud service combination, weapon system combination or portfolio and the like under complicated and incomplete information are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of Internet service combination, and in particular relates to a service combination optimization method under incomplete information. Background Art

[0002] Combinatorial optimization problems are very representative of engineering. Many problems in various fields such as industry, manufacturing, communications, transportation, military, and economy can be abstracted into combinatorial optimization problems, such as packing problems, cloud service combinations, weapon system combinations, and investment portfolios. Combinatorial optimization problems are usually multi-objective optimization problems under complex constraints and are difficult to solve. The main reason is that the algorithms for solving these problems require extremely long running time and huge storage space, which makes it impossible to implement them on existing computers, the so-called combinatorial explosion.

[0003] In the past, combinatorial optimization solutions included branch-and-bound and dynamic programming. Currently, the mainstream algorithms are heuristic algorithms, such as genetic algorithms, ant colony algorithms, and sparrow algorithms. These methods are suitable for solving problems under complete information. In reality, combinatorial optimization is to find the optimal solution from the feasible solution set of multi-objective service combination optimization problems while considering many objectives and decision makers' preferences. Since it is difficult to obtain reliable estimates of future service combination results and extract preferences, service combination decisions are usually made under incomplete information. In this case, the purpose of the solution is to narrow the scope of service combinations and obtain a set of non-dominated combinations, rather than to provide an optimal service combination. When the number of services is large, the number of all possible service combinations will be huge, which will result in the number of non-dominated combinations being very high and impossible to solve in a reasonable time using an accurate algorithm. Summary of the invention

[0004] The purpose of the present invention is to solve the problem of difficulty in obtaining the optimal service combination under incomplete information, and to achieve the problem of algorithm accuracy and speed in the case of huge service combinations.

[0005] To achieve the above object, the present invention provides a service combination optimization method under incomplete information, and the specific steps are as follows: S1. Data initialization; initialization parameters include non-dominated sets , calculation times N, Set to the empty set.

[0006] S2. Feasible indicators for each service goal It is incomplete information, that is, it generates random feasible indicators for each target of the service The target indicator vector v that constitutes the service portfolio, is the element in the lower bound matrix, is the element in the upper bound matrix, where ,and is a random number. A random number that follows a uniform distribution Multiply it by the difference between the upper and lower bounds of the indicator, and add the indicator generated by the lower bound of the indicator as a random indicator, that is, ,in Compliance interval Uniform distribution, that is .

[0007] S3. Generate random feasible weight vector As the extreme point of the feasible interval of weight combination.

[0008] S301, generating extreme points of convex weight space .

[0009] S302: Generate t service index distributions Random number ; Normalized random number , get the random amount ,in = .

[0010] S303, random feasible weight vector Each element in the formula Calculate and obtain, is the weight, It follows an exponential distribution.

[0011] S4. Solve the 0-1 integer programming problem to obtain the optimal value

[0012]

[0013] in , is the decision vector, is the first element, m is the number of feasible indicators of the target, and T is the transposition; and are the constraint matrix and constraint vector respectively. The elements of the constraint matrix A For interval uniformly distributed integers, where q is the number of constraints. Each element of the constraint vector B Set to 50% of.

[0014] S5. Add the optimal value obtained in step S4 to the set middle.

[0015] S6: Whether the set number of times N is reached, if not, repeat steps S2-S5, otherwise output the set .

[0016] S7, pair set The elements are tested for non-domination, and then the optimal service combination is obtained using the pessimistic criterion, the optimistic criterion, the optimistic coefficient method, or the regret value minimization and maximization criterion; A service composition optimization method under incomplete information is as follows: S1. Initialization, setting non-dominated sets It is an empty set and the number of calculations is N.

[0017] S2. Generate a random feasible weight vector As the extreme point of the feasible interval of weight combination.

[0018] S201, generating extreme points in convex weight space .

[0019] S202. Generate t service index distributions Random number ; Normalized random number , get the random amount ,in = .

[0020] S203, random feasible weight vector Each element in the formula Calculate and obtain, is the weight, It follows an exponential distribution.

[0021] S3. Calculate the ideal vector . Calculate the target indicators of all service combinations and determine the upper bound of each target indicator. Randomly generate the ideal vector , requiring that no element in the vector is inferior to the upper bound of each target indicator.

[0022] S4. Construct random vector . The elements in follow an exponential distribution and are unit vectors. They are generated from the exponential distribution Exp(1) Random numbers , a random vector Elements , thus constructing the vector .

[0023] S5. Generate random feasible indicators for each service target , which constitutes the target indicator vector v of the service combination. Construct the setting set X, and randomly select elements from the set X to form a subset , randomly select an element from X , if the element belongs to the set X, then the index vector v element takes the upper bound of the index , otherwise it is the lower bound of the indicator ,Right now ,but ,if ,but .

[0024] S6. Solve the 0-1 integer programming problem:

[0025]

[0026]

[0027] in is a real variable, is the i-th element in the random vector, is the i-th element of the ideal vector, is the set of real numbers, is the decision vector elements, is the extreme point of the i-th weight feasible interval; , is the decision vector, is the first element, m is the number of feasible target indicators, T is the transposition, and are the constraint matrix and constraint vector respectively. The elements of the constraint matrix A For interval Uniformly distributed integers, where is the number of constraints; each element of the constraint vector B Set to 50%; get the optimal value of the above problem and add it to the set middle.

[0028] S7, whether the set number of times N is reached, if not, repeat steps S2-S6, otherwise output the set .

[0029] S8. Pair collection The elements are tested for non-domination, and then the pessimistic criterion, optimistic criterion, optimistic coefficient method or regret value minimization and maximization criterion are used to obtain the optimal service combination.

[0030] Beneficial effects: Large-scale combinatorial optimization problems are impractical or costly to accurately solve due to high computational complexity or resource limitations. Method 1 and method 2 provided by the present invention can quickly search for approximate solutions to combinatorial optimization, reduce computing time, and reduce the use of computing resources. The balance between the accuracy of the solution and the computing cost can be controlled by adjusting the algorithm parameters to adapt to different application scenarios. It has strong flexibility and is particularly suitable for application scenarios that require rapid response. Compared with the enumeration comparison method, the method provided by the present invention can obtain an approximate optimal solution on the basis of ensuring accuracy, and is far less than the enumeration comparison method in terms of computing time and computing resources. The present application solves the problems of slow speed of solving combinatorial optimization under incomplete information and accuracy of approximate algorithms through the above method, and can be widely used in the field of engineering, especially in complex and incomplete information packaging problems, cloud service combinations, weapon system combinations or investment portfolios. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a flow chart of method 1 provided in an embodiment of the present invention; Figure 2 is a flow chart of method 1 provided in an embodiment of the present invention; Figure 3 It is a simulation result diagram of method 1 and method 2 provided by the present invention; Figure 4 It is a tourism service combination evaluation system provided by the embodiment of the present invention; Figure 5 is the total trust loss of the embodiment provided by the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the 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.

[0033] The application principle of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0034] Embodiment 1: Method 1, such as Figure 1 As shown in Figure 1, a service composition optimization method under incomplete information is as follows: S1. Data initialization; initialization parameters include non-dominated sets , calculation times N, Set to the empty set.

[0035] S2. Feasible indicators for each service goal It is incomplete information, that is, it generates random feasible indicators for each target of the service The target indicator vector v that constitutes the service portfolio, is the element in the lower bound matrix, is the element in the upper bound matrix, where ,and is a random number. A random number that follows a uniform distribution Multiply it by the difference between the upper and lower bounds of the indicator, and add the indicator generated by the lower bound of the indicator as a random indicator, that is, ,in Compliance interval Uniform distribution, that is .

[0036] S3. Generate random feasible weight vector As the extreme point of the feasible interval of weight The combination is as follows: S301, generating extreme points of convex weight space .

[0037] S302: Generate t service index distributions Random number ; Normalized random number , get the random amount ,in = .

[0038] S303, random feasible weight vector Each element in the formula Calculate and obtain, is the weight, It follows an exponential distribution.

[0039] S4. Solve the 0-1 integer programming problem to obtain the optimal value

[0040]

[0041] in , is the decision vector, is the first element, m is the number of feasible indicators of the target, and T is the transposition; and are the constraint matrix and constraint vector respectively. The elements of the constraint matrix A For interval uniformly distributed integers, where q is the number of constraints. Each element of the constraint vector B Set to 50% of.

[0042] S5. Add the optimal value obtained in step S4 to the set middle.

[0043] S6: Whether the set number of times N is reached, if not, repeat steps S2-S5, otherwise output the set .

[0044] S7, pair set The elements are tested for non-domination, and then the pessimistic criterion, optimistic criterion, optimistic coefficient method or regret value minimization and maximization criterion are used to obtain the optimal service combination.

[0045] Method 2, such as Figure 2 As shown in Figure 1, a service composition optimization method under incomplete information is as follows: S1. Initialization, setting non-dominated sets It is an empty set and the number of calculations is N; S2. Generate a random feasible weight vector As the extreme point of the feasible interval of weight combination of; S201, generating extreme points in convex weight space ; S202. Generate t service index distributions Random number ; Normalized random number , get the random amount ,in = ; S203, random feasible weight vector Each element in the formula Calculate and obtain, is the weight, It follows an exponential distribution.

[0046] S3. Calculate the ideal vector ; Calculate the target indicators of all service combinations and determine the upper bound of each target indicator; Randomly generate ideal vectors , requiring that no element in the vector is inferior to the upper bound of each target indicator; S4. Construct random vector ; The elements in follow an exponential distribution and are unit vectors; they are generated from the exponential distribution Exp(1) Random numbers , a random vector Elements , thus constructing the vector ; S5. Generate random feasible indicators for each service target , forming the target indicator vector v of the service combination; constructing the setting set X, randomly selecting elements from the set X to form a subset , randomly select an element from X , if the element belongs to the set X, then the index vector v element takes the upper bound of the index , otherwise it is the lower bound of the indicator ,Right now ,but ,if ,but ; S6. Solve the 0-1 integer programming problem:

[0047]

[0048]

[0049] in is a real variable, is the i-th element in the random vector, is the i-th element of the ideal vector, is the set of real numbers, is the decision vector elements, is the extreme point of the i-th weight feasible interval; , is the decision vector, m is the number of feasible target indicators, T is the transposition, and are the constraint matrix and constraint vector respectively. The elements of the constraint matrix A For interval Uniformly distributed integers, where is the number of constraints; each element of the constraint vector B Set to 50%; get the optimal value of the above problem and add it to the set middle; S7, whether the set number of times N is reached, if not, repeat steps S2-S6, otherwise output the set ; S8. Pair collection The elements are tested for non-domination, and then the pessimistic criterion, optimistic criterion, optimistic coefficient method or regret value minimization and maximization criterion are used to obtain the optimal service combination.

[0050] The pessimistic criterion in step S7 of method 1 and step S8 of method 2 means that in the decision-making process, the decision maker holds the most pessimistic attitude, always estimates things very unfavorably, and then chooses the plan that can provide the maximum reward in the worst case. The optimistic criterion assumes that the decision maker is optimistic about the future results, always assumes that the most favorable state for himself has appeared, and believes that the best choice is to choose the most favorable result among the most favorable results. The optimistic coefficient method is that the decision maker can focus on making decisions between excessive optimism and excessive pessimism, neither completely optimistic nor completely pessimistic, but taking a certain proportion between the two, thereby forming a new discriminant value and taking the plan with the maximum value. The regret value minimization maximization criterion After choosing a certain plan, if the natural state that occurs in the future shows that other plans have greater benefits, the decision maker will regret his choice. Therefore, the goal of this method is to make the regret value as small as possible.

[0051] Embodiment 2: In the service composition model, assuming there are m services, evaluated by n criteria, the service set can be expressed as ,project No. Standard indicators are used Therefore, the multi-attribute value function theory can be used to obtain the project The overall indicators are:

[0052] (1) is the weight of the ith criterion. Assume a project portfolio For a subset of all projects, the set of all feasible service combinations can be the power set The total indicators of the service portfolio are:

[0053] (2) in ,if ,but . is the target indicator vector, is a random feasible weight vector. The service composition is also constrained by the available resources, and these constraints can be modeled by a set of linear inequalities. Assuming the number of constraints is q, and are constraint matrices and constraint vectors respectively, then the set of feasible service combinations for:

[0054] (3) If the indicators and weights are fully informed, the optimal portfolio is obtained by optimizing formula (3) in the set of feasible portfolios, that is, the optimal investor can be regarded as the solution of the integer linear programming problem: (4) in is the decision vector The key feature of the service portfolio model is that decision makers can only provide incomplete information about weights and indicators. Weights and indicators can be replaced by interval estimates instead of point estimates. and feasible weights Set as: and

[0055] For each portfolio , the total index is also expressed in intervals, that is, (5) The effect of the method of the present invention is illustrated by adopting a problem with 6 objectives, 3 resource constraints and 100,000 combinations.

[0056] The upper and lower bound matrices of the problem are and , the indicator upper limit matrix Elements in For interval The service is uniformly distributed integers, and the elements in the index lower limit matrix are ,in .

[0057] Constraint matrix A and constraint vector B, elements of constraint matrix A For interval uniformly distributed integers, where q is the number of constraints. Each element of the constraint vector B Set to 50% of.

[0058] There is no weight information, that is Method 1, method 2 and the enumeration method of the present invention randomly search for non-dominated investment portfolios. Figure 3 and Table 1. Figure 3It can be seen from the figure that method 1 searches for more service combinations than method 2 at the beginning. As the number of iterations increases, the speed of searching for new service combinations gradually decreases from method 2, and the search speeds of both method 1 and method 2 are greater than the enumeration method. Therefore, after enough approximation rounds, the speed at which method 2 finds common service combinations is approximately linear, and is also close to the speed of method 1. After 1000 simulations, the results of the method core index are shown in Table 1. Obviously, the number of investment portfolios generated by method 2 is more than that of method 1, and the accuracy is also higher than that of method 1. The core set, edge set, and external set of the two methods are similar in the final stage. In the same time, the two methods search for more non-dominated investment portfolios than the enumeration method, and the accuracy is the second highest than the enumeration method. In order to improve the accuracy, the preferred solution is to perform a non-dominated test on the non-dominated investment portfolio, and then use the pessimistic criterion or the optimistic criterion or the optimistic coefficient method or the regret value minimization maximization criterion to obtain the optimal service portfolio.

[0059] Table 1 Comparison of the results of various methods after 1000 simulations

[0060] Embodiment 3: The use of this method is illustrated by solving the tourism cloud service combination under incomplete information. The tourism cloud service combination usually covers a series of interrelated and synergistic elements to meet the various needs of tourists during the travel process. The general process is: the user has a travel demand, searches through the Internet, and then makes a reservation, and begins to accept the route arrangement, transportation, accommodation, catering, tour guide and other sub-services from the service provider, and finally gives an evaluation of each sub-service. It is difficult for a service provider to provide all tourism services to meet the requirements of tourists. Usually, a team of several service providers is formed to complete the tourism service together, which can not only meet the service requirements of tourists, but also provide personalized services. The advantages and disadvantages of cloud service combinations are distinguished by evaluating the trust of each sub-service. When the number of service providers is large, it takes a long time to find the optimal service combination, which cannot meet the time requirements of tourism cloud service combination. This method can be used to search for approximate service combinations, which can meet the time requirements of service combination solution.

[0061] The tourism service portfolio trust evaluation system is described in Figure 4The entire tourism service is evaluated from five sub-services: route arrangement, transportation, tour guide, accommodation and catering. Each sub-service is described by 3-6 attributes. Transportation services include punctuality and comfort. Catering services can be described by six aspects: restaurant location, waiting time for dining, price, hygiene, taste and dining environment. Accommodation services include hotel location, hygiene, hotel facilities, price and hotel surroundings. Tour guide services can be evaluated from the guide's service attitude, professional knowledge, explanation ability, price and equipment conditions. Route arrangement services include the popularity of scenic spots, distance, compactness of itinerary, price and density of people in the route.

[0062] Assume a tourism service scenario where 12 out-of-town tourists form a group and want to go to Xi'an for a one-day tour. They make a request through a service agent. According to the needs of tourists, the service agent organizes existing service providers into teams to form several service combinations. The trust evaluation of tourism cloud service combinations uses an additive value function. Since tourists often cannot accurately give their preferences, that is, the preferences of tourists are uncertain, the RICH method is used to model preferences (weights). Attribute values ​​are represented by interval values. The data comes from Ctrip.com, Qunar.com, Mafeng.com and Meituan.com. 15 route service providers, 20 catering service providers, 20 accommodation service providers, 20 tour guide service providers, and 10 transportation service providers are selected from them. The number of tourism service combinations is 1,200,000. Except for the distance, location and price attributes, the remaining attribute values ​​are derived from the evaluation of the service requester, and the data is normalized. The preferences of passengers are as follows: they think that route arrangement is the most important, followed by accommodation, then catering, and finally tour guides and transportation. Among the attributes of route arrangement, popularity is the most important, followed by the compactness of the itinerary, then the density of people, then the price, and finally the total distance to the attractions. Among the attributes of accommodation, the most important is hygiene, followed by location, then facilities, and finally environment and price. Among the attributes of catering, the most important is hygiene, followed by taste, then waiting time, then price, and finally location and environment. Among the attributes of tour guides, the most important is service attitude, followed by the level of explanation, then price, and finally equipment conditions and professional knowledge. Among the attributes of transportation, punctuality is more important, followed by comfort, and finally price.

[0063] The simulation platform uses a computer with CPUi5-11400, 16G memory, Windows 10 Professional operating system, and MATLAB2014 software. Using method 1 of the present invention to generate 1000 non-dominated combinations, the average total trust loss of the optimal tourism cloud service combination is 0.098, and the calculation time is 47 seconds. The average total trust loss of method 2 is 0.091, and the calculation time is 36 seconds. 5000 non-dominated combinations are generated. The total average trust loss of method 1 is 0.057, and the calculation time is 461 seconds. The total average trust loss of method 2 is 0.052, and the calculation time is 204 seconds. The specific total trust loss is shown in Figure 5 The calculation time of generating 1000 non-dominated combinations by enumeration method is 197 seconds, and the time of obtaining the optimal tourism combination by enumeration method is 3 hours and 37 minutes. The method of the present invention solves the tourism cloud service combination problem well and meets the time requirement.

[0064] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A service composition optimization method under incomplete information, characterized in that: The following steps are involved: S1. Data initialization, initialization parameter non-dominated set It is an empty set and the number of calculations is N; S2. Feasible indicators for each service goal It is incomplete information, that is, it generates random feasible indicators for each target of the service The target indicator vector v that constitutes the service portfolio, is the element in the lower bound matrix, are the elements in the upper bound matrix, where ,and is a random number; a random number that obeys a uniform distribution Multiply it by the difference between the upper and lower bounds of the indicator, and add the indicator generated by the lower bound of the indicator as a random indicator, that is, ,in Compliance interval Uniform distribution, that is ; S3. Generate random feasible weight vector As the extreme point of the feasible interval of weight combination of; S4. Solve the 0-1 integer programming problem to obtain the optimal value ； ； in , is the decision vector, is the first element, m is the number of feasible indicators of the target, and T is the transposition; and are the constraint matrix and constraint vector respectively. The elements of the constraint matrix A For interval Uniformly distributed integers, where is the number of constraints; each element of the constraint vector B Set to 50%; S5. Add the optimal value obtained in step S4 to the set middle; S6: Whether the set number of times N is reached, if not, repeat steps S2-S5, otherwise output the set ; S7, pair set The elements are tested for non-domination, and then the pessimistic criterion, optimistic criterion, optimistic coefficient method or regret value minimization and maximization criterion are used to obtain the optimal service combination.

2. The service composition optimization method under incomplete information according to claim 1, characterized in that: Step S3 is as follows: S301, generating extreme points of convex weight space ; S302: Generate t service index distributions Random number ; Normalized random numbers , get the random amount ,in = ; S303, random feasible weight vector Each element in the formula Calculate and obtain, is the weight, It follows an exponential distribution.

3. A service composition optimization method under incomplete information, characterized in that: The following steps are involved: S1. Initialization, setting non-dominated sets It is an empty set and the number of calculations is N; S2. Generate a random feasible weight vector As the extreme point of the feasible interval of weight combination of; S3. Calculate the ideal vector ; Calculate the target indicators of all service combinations and determine the upper bound of each target indicator; Randomly generate ideal vectors , requiring that no element in the vector is inferior to the upper bound of each target indicator; S4. Construct random vector ; The elements in follow an exponential distribution and are unit vectors; they are generated from the exponential distribution Exp(1) Random numbers , a random vector Elements , thus constructing the vector ; S5. Generate random feasible indicators for each service target , forming the target indicator vector v of the service combination; constructing the setting set X, randomly selecting elements from the set X to form a subset , randomly select an element from X , if the element belongs to the set X, then the index vector v element takes the upper bound of the index , otherwise it is the lower bound of the indicator ,Right now ,but ,if ,but ; S6. Solve the 0-1 integer programming problem: ； ； ； in is a real variable, is the i-th element in the random vector, is the i-th element of the ideal vector, is the set of real numbers, is the decision vector elements, is the extreme point of the i-th weight feasible interval; , is the decision vector, is the first element, m is the number of feasible target indicators, T is the transposition, and are the constraint matrix and constraint vector respectively. The elements of the constraint matrix A For interval Uniformly distributed integers, where is the number of constraints; each element of the constraint vector B Set to 50%; get the optimal value of the above problem and add it to the set middle; S7, whether the set number of times N is reached, if not, repeat steps S2-S6, otherwise output the set ; S8. Pair collection The elements are tested for non-domination, and then the pessimistic criterion, optimistic criterion, optimistic coefficient method or regret value minimization and maximization criterion are used to obtain the optimal service combination.

4. The service composition optimization method under incomplete information according to claim 3 is characterized in that: Step S2 is specifically as follows: S201, generating extreme points in convex weight space ; S202. Generate t service index distributions Random number ; Normalized random numbers , get the random amount ,in = ; S203, random feasible weight vector Each element in the formula Calculate and obtain, is the weight, It follows an exponential distribution.

Images