A modular service configuration optimization method considering multi-user and fuzzy house of quality information

Abstract

This paper proposes a modular service configuration optimization method that considers multiple users and fuzzy house of quality information. The method includes the following steps: 1. Establishing a modular service structure using a design structure matrix to represent complex service processes; 2. Constructing a two-level quality factored house of quality; 3. Establishing a multi-client service configuration optimization model that comprehensively considers service process time constraints, service component configuration constraints, budget constraints, and service duration constraints; 4. Improving the established multi-objective service configuration optimization model into a linear programming model; 5. Expanding the model based on uncertainty in the house of quality and further converting it into a three-objective model; 6. Constructing recommended solution sets for service configuration optimization problems of different scales using different algorithms; 7. Using a K-means clustering method to reduce the solution set to a set of candidate solutions, and finally allowing customers to adjust their goals and interactively select satisfactory solutions. This method has great practical significance for both service companies and customers.

Description

Technical Field

[0001] The present invention relates to the field of information technology and automation technology, and in particular to a modular service configuration optimization method considering multi-user and fuzzy quality house information. Background Art

[0002] Service configuration involves combining modular service components within a service module in a specific manner to create a service solution that meets specific customer needs. During modular service configuration, service components within different modules face complex configuration constraints and may also face sequential execution relationships, such as parallel and serial execution. The selection and combination of different service components, as well as the required support for various service resources, directly impacts service completion time, budget, and quality, ultimately leading to unfeasible configuration outcomes such as extended schedules, budget overruns, and reduced quality. Therefore, incorporating service process information into service configuration is crucial for developing diverse and feasible service solutions.

[0003] The process of the existing service configuration method is as follows: the service entity analyzes the service characteristics, decomposes the service into several modules, and establishes corresponding service interfaces to ensure that the services can be connected and carried out smoothly; then, different service components are set for different service modules, such as the common setting of high, medium and low-level service components to form a service module; next, the customer can make requirements on the service cycle, budget, demand, etc.; finally, the service entity conducts a combination analysis based on the attribute values ​​of each service component, recommends a service plan that meets the requirements to the customer, and then forms a letter of intent.

[0004] In traditional service configuration optimization, service components exist relatively independently. However, due to the temporal relationships between service modules, which can be parallelized or serialized, and the resource constraints of service entities, the service configuration process is subject to critical characteristics. Therefore, simply considering simple service combinations during service configuration can lead to drawbacks such as excessively long service cycles, increased service budgets, and poor service performance. Analysis of numerous service examples shows that some modules can be performed in parallel, such as waterway and electrical system renovations, while others require serialization, such as waterway and waterproofing projects. Furthermore, when service companies implement multiple projects simultaneously, the effective scheduling of service human and material resources significantly impacts service duration, budget, and performance. Summary of the Invention

[0005] To address the problems of the above-mentioned prior art, the present invention proposes a modular service configuration optimization method that considers multi-user and fuzzy house of quality information. On the basis of taking into account the service configuration combination, when the service company recommends service solutions to customers, it considers customer budget, duration, performance and other qualities as well as the uncertainty of the house of quality, improves the service configuration optimization calculation method, incorporates the time constraints of the service process, the configuration constraints of the service components, the budget constraints and the service duration, and reasonably optimizes the solutions provided to customers.

[0006] To achieve the above technical objectives, the technical solution adopted by the present invention is a modular service configuration optimization method considering multi-user and fuzzy quality house information, comprising the following steps:

[0007] Step 1: Establish a service modular structure that uses the Design Structure Matrix (DSM) method to represent the complex service process, and define the service components available in the service module and their basic attribute values, such as cost and time;

[0008] Step 2: Construct a two-level QFD (Quality Function Deployment) quality house to map the importance of different customer needs to the service component preferences (performance index) of each service module;

[0009] Step 3: Establish a multi-customer service configuration optimization model that comprehensively considers the time constraints of the service process, the configuration constraints of service components, the budget constraints, and the service duration constraints;

[0010] Step 4: Improve the established multi-objective service configuration optimization model into a linear programming model;

[0011] Step 5: Based on the uncertainty of the quality house, the model is expanded and further transformed into a three-objective model;

[0012] Step 6: Construct a set of recommended solutions for different algorithms to calculate service configuration optimization problems of different scales;

[0013] Step 7: Construct a K-means clustering method to reduce the solution set and form a candidate solution set. Finally, the customer will make target adjustments and interactive selections for satisfactory solutions.

[0014] Furthermore, in step 2, the importance of different customer requirements is mapped to the service component preferences of each service module. First, a service planning house of quality is established, where the "left wall" represents customer requirements and the "ceiling" represents basic service attributes. Then, a module configuration house of quality is established, where the "left wall" represents basic service attributes and the "ceiling" represents the service components of each service module. This two-level service house of quality allows the importance of customer requirements to be mapped to the service component preferences of each service module.

[0015] Furthermore, in step 3, in the configuration service optimization model considering the service process information, the optimization objectives are obj1 and obj2; the influencing constraints are a1, a2, a3, a4, a5, a6, a7, a8, and a9. The detailed parameter information and model are as follows:

[0016] (1) Model parameter information:

[0017] N m : The number of required service modules;

[0018] N i : Service module component, i=1, 2, ..., N m ;

[0019] d ij : Service processing time of service module component, i=1, 2, ..., N m ; j = 1, 2, ..., N i ;

[0020] c ij : Service unit cost of service module component, i = 1, 2, ..., N m ; j = 1, 2, ..., N i ;

[0021] C fix : Fixed fee for services;

[0022] H: Number of customers who need to configure the service

[0023] C h : The upper limit of the total cost budget required by the customer, h = 1, 2, ..., H;

[0024] T h : The upper limit of the total service time required by the customer, h = 1, 2, ..., H;

[0025] K1: The number of customer demands;

[0026] K2: The number of service attributes is;

[0027] w k1 : The relative importance of customer needs k1 = 1, 2, ..., K1;

[0028] q1 k1k2 : The correlation between customer needs and service attributes, k1 = 1, 2, ..., K1; k2 = 1, 2, ..., K2;

[0029] q2 k2ij: The relationship between service attributes and service module components k2 = 1, 2, ..., K2; i = 1, 2, ..., N m ; j = 1, 2, ..., N i ;

[0030] The temporal relationship of activities in the service process, Indicates that the i-th service module is the predecessor of the i′-th service module, otherwise i, i′=1, 2, ..., N m ;i≠i′;

[0031] s hi : The start time of the service module i = 1, 2, ..., N m , h=1,2,...,H;

[0032] T: maximum service period, integer;

[0033] L: the number of resource types, integer;

[0034] R lt : the total amount of the lth resource at time t, l = 1, 2, ..., L; t = 1, 2, ..., T;

[0035] r ijl : The resources required by the jth module component of the i-th service module during execution i=1, 2, ..., N m ; j = 1, 2, ..., N i ; l = 1, 2, ..., L;

[0036] δ iji′j′ : Compatibility relationship of service module components, i, i′=1, 2, ..., N m ; j, j′ = 1, 2, ..., N i ;i&j≠i′&j′. When there is a compatibility relationship between module components (if the j′th module component of the i′th service module is selected in the configuration scheme, the jth module component of the i′th service module must also be selected), δ iji′j′ =1, otherwise δ iji′j′ =0;

[0037] δ′ iji′j′ : Exclusive relationship of service module components, i, i′=1, 2, ..., N m ; j, j′ = 1, 2, ..., N i ; i&j≠i′&j′. When there is an exclusive relationship between module components (if the j′th module component of the i′th service module is selected in the configuration scheme, the jth module component of the i′th service module cannot be selected), δ′iji′j′ =1; otherwise δ′ iji′j′ =0;

[0038] x hij : Main decision variable, i = 1, 2, ..., N m ; j = 1, 2, ..., N i , h=1, 2, ..., H. If the jth module component of the i-th service module is selected in the configuration scheme, then x hij =1; otherwise x hij =0.

[0039] y hijt : The auxiliary variable represents the task end status of the jth module component of the i-th service module at time block t, that is, if time block t is the end time of the module component, then y hijt =1; otherwise y hijt =0.

[0040] (2) The multiple targets to be optimized include obj1 and obj2, where:

[0041] Optimization objective obj1: Maximize the minimum value of the customer satisfaction index (CSI) of each customer, expressed as the formula:

[0042]

[0043] Optimization objective obj2: maximize R, expressed as formula:

[0044]

[0045] (3) The impact constraints considered are a1, a2, a3, a4, a5, a6, a7, and a8, where:

[0046] Impact constraint a1: Only one module component in each service module can be selected. This relationship can be expressed as:

[0047]

[0048] Impact constraint a2: The temporal relationship of service module nodes in the service process. This relationship can be expressed as:

[0049]

[0050] Impact constraint a3: The compatible and exclusive configuration relationship between service module components. This relationship can be expressed as:

[0051] x hij ≥δ iji′j′ x hi′j′ i, i′=1, 2, ..., Nm ; j, j′ = 1, 2, ..., N i ;i&j≠i′&j′;h=1,2,...,H

[0052] x hij ≤2-δ′ iji′j′ -x hi′j′ i, i′=1, 2, ..., N m ; j, j′ = 1, 2, ..., N i ;i&j≠i′&j′;h=1,2,...,H

[0053] Impact constraint a4: During the entire service process, the resources occupied by each module component cannot exceed its upper limit. This relationship can be expressed as:

[0054]

[0055] Influence constraint a5: limits the auxiliary variable y ijt The value of conforms to its definition, and this relationship can be expressed as:

[0056]

[0057]

[0058] Impact constraint a6: The total cost and total duration of the service cannot exceed the upper limit given by the customer. This relationship can be expressed as:

[0059]

[0060]

[0061] Impact constraint a7: The initial setting of the service component start time and the non-negative constraint. This relationship can be expressed as:

[0062] s h1 =0,s hi ≥0 i=2,...,N m +1; h=1, 2, ..., H

[0063] Influence constraint a8: gives the specific definition of the configuration decision variables and restricts the decision variables. This relationship can be expressed as:

[0064] x ij =0 or 1 i=1, 2, ..., N m ; j = 1, 2, ..., N i ; h=1, 2, ..., H

[0065] Furthermore, step 4 is specifically as follows:

[0066] By analyzing the multi-objective optimization model (P), we know that constraint a4 is a nonlinear constraint, but the decision variables in this constraint are all 0-1 variables, which can be easily converted into equivalent linear constraints. Define a new auxiliary decision variable z hijt i=1,2,...,N m ; j = 1, 2, ..., N i ; t=1,2,…,T;h=1,2,…,H, replace constraint a4 in the model with the following constraint:

[0067]

[0068] M(2-x hij -y hijt )≥1-z hijt i=1,2,...,N m ; j = 1, 2, ..., N i ;t=1,2,...,T;h=1,2,...,H

[0069] (a4-2)

[0070] z hijt ≤x hij , z hijt ≤y hijt i=1,2,...,N m ; j = 1, 2, ..., N i ;t=1,2,...,T;h=1,2,...,H

[0071] (a4-3)

[0072] z hijt =0 or 1

[0073] (a4-4)

[0074] After the conversion, the decision variables of the multi-objective linear optimization model are 0-1 quantities, the coefficients can be normalized into integer type parameters, and all Pareto solutions can be obtained using an exact algorithm.

[0075] Furthermore, in step 5, QFD is a method for integrating quantitative and qualitative information, some of which is obtained by human experts based on their personal experience and has inherent uncertainty. Since there is no historical data for estimating the elements of HoQ1 and HoQ2, which represent a subjective uncertainty, it is best to represent them as fuzzy numbers and assume that the relationship between customer needs and service attributes in HoQ1 and HoQ2 is a triangular fuzzy number, where the most pessimistic values ​​are: Most likely value: Most optimistic value: Consider the optimization model based on fuzzy QFD, where the optimization objectives are obj3, obj4, and obj5; the influencing constraints are, where the detailed parameter information and model are as follows

[0076] (1) Model parameter information:

[0077] The relationship between customer needs and service attributes in HoQ1 is a triangular fuzzy number, expressed as:

[0078]

[0079] The relationship between customer needs and service attributes in HoQ2 is a triangular fuzzy number, expressed as:

[0080]

[0081] (2) Required optimization objectives

[0082] Obj3: The optimization model formula based on fuzzy HOQ is as follows:

[0083]

[0084] Among them I h It is not a triangular fuzzy number, but an LR type fuzzy number. h It can be expressed as

[0085] The impact constraints considered are:

[0086] Most pessimistic value

[0087] Most likely value

[0088] Most optimistic value

[0089] Obj4: Based on Theorem 1 and Proposition 1, Model III is restated as follows:

[0090] Max I *

[0091] The impact constraints considered are:

[0092] a12

[0093]

[0094] (The rest of the constraints are the same as the basic model)

[0095] Obj5: The three-objective optimization model formula is as follows:

[0096]

[0097]

[0098]

[0099] The impact constraints considered are:

[0100]

[0101]

[0102]

[0103]

[0104] (The rest of the constraints are the same as the basic model)

[0105] in, All are continuous decision variables.

[0106] Furthermore, in step 6, for small- and medium-scale service configuration problems, the present invention designs an objective function value space partitioning algorithm for optimization, converting the multi-objective optimization problem into a single-objective optimization problem. Each time the single-objective linear integer programming problem is solved, a new non-dominated solution is obtained. The search space is then partitioned using the new non-dominated solution, the space dominated by the non-dominated solution is eliminated, and the non-dominated solution set is updated. This process is repeated until no new non-dominated solutions are generated, resulting in the complete Pareto solution set for the multi-objective optimization problem. When solving the single-objective linear integer programming model for each subproblem, optimization software such as CPLEX can be used for calculation.

[0107] For large-scale service configuration problems, this paper designs the NSGA-II algorithm for optimization. Based on NSGA, it performs a fast non-dominated sort and introduces a congestion comparison operator and an elitist retention strategy, significantly reducing the algorithm's computational complexity while significantly improving its performance. The chromosomes used to solve the service configuration model in this paper primarily employ an integer encoding structure, model constraints are handled using a penalty function method, and genetic operations employ single-point crossover and neighborhood mutation.

[0108] Furthermore, compared to existing methods, the service configuration method proposed in step 7 can simultaneously generate recommended configuration solutions for multiple users and establish a fuzzy QFD-based optimization model for the configuration problem. This model is then converted into a three-objective model, and both exact and metaheuristic algorithms are designed for small-scale and large-scale problems, respectively. The effectiveness of both algorithms is demonstrated through numerical experiments and industrial examples. Therefore, this invention can help businesses provide personalized and diversified configuration services to customers with individual needs, while also improving customer satisfaction and increasing a company's potential market share, thus possessing significant practical significance.

[0109] Description of the accompanying tables and figures

[0110] Figure 1 It is the modular structure and configuration of the service described in the present invention;

[0111] Figure 2 Schematic diagram of the calculation steps of the method of the present invention;

[0112] Figure 3 It is the timing relationship between specific modules in the present invention;

[0113] Figure 4 It is a two-level quality house mapping diagram of customer needs in the specific implementation process of the present invention;

[0114] Figure 5 is the budget sensitivity analysis result in the specific service configuration case of the present invention;

[0115] Figure 6 It is the result of the service term sensitivity analysis experiment in the specific service configuration case of the present invention;

[0116] Figure 7 is a main effect diagram of the parameters determined in the specific service configuration case of the present invention; wherein, Figure 7 (a) is the main effect plot of mean GD, Figure 7 (b) is the main effect diagram of mean time. DETAILED DESCRIPTION

[0117] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following specific embodiments further illustrate the present invention in detail.

[0118] First, if Figure 1As shown in the figure, to meet the diverse needs of customers, the service is designed into a modular structure for customers to choose from, where the service modules can be divided into mandatory modules M1 and optional modules M2. The service process can be described by the service process, in which complex temporal relationships may exist between service modules. Some service modules require the completion of other service modules before they can be carried out, that is, the order relationship between service modules. During the configuration process, compatible and exclusive relationships may exist between service module components. Customer demand preference information and demand constraint information are the main inputs of the service configuration problem, and different customers have different specific requirements. With the support of the service modular structure, considering the temporal constraints of the service process and the configuration constraints between service components, different optimization objectives can be selected to establish an optimization model.

[0119] Secondly, if Figure 2 As shown, a specific process of a service configuration optimization method considering service process information includes the following steps:

[0120] Step 1: Establish a modular service structure. By analyzing service characteristics, find decomposable service elements and form N m Develop a service module and analyze the connections between service elements to establish a service interface and modular structure. For complex service processes, the Design Structure Matrix (DSM) approach can be used to obtain service modules with high cohesion and low coupling. Define the service components available in the service module and their basic attribute values, such as cost and time.

[0121] Step 2: Construct a two-level QFD (Quality Function Deployment) House of Quality. Drawing on the House of Quality structure used in product quality design in QFD, first establish the Service Planning House of Quality, where the "left wall" represents customer needs and the "ceiling" represents basic service attributes. Next, establish the Module Configuration House of Quality, where the "left wall" represents basic service attributes and the "ceiling" represents the service components of each service module. This two-level service House of Quality allows the importance of customer needs to be mapped to the service component preferences of each service module.

[0122] By drawing on the first two levels of the four-level quality house structure of QFD product quality design, we successively built the service planning quality house and the module configuration quality house, mapping the importance of customer needs to the service components of the service module.

[0123] (1) First, analyze and clarify customer needs. Different customers rate the importance of all needs based on their own needs, and normalize the importance of different needs;

[0124] (2) Secondly, the decoration experts will identify and evaluate the mapping relationship between the two quality houses, and the values ​​of the quality house parameters will be obtained by comprehensive analysis by relevant experts convened by the enterprise.

[0125] (3) Finally, the corresponding performance value is calculated based on the module configuration quality house.

[0126] Step 3: Establish a multi-objective service configuration optimization model (P). Based on the above two steps, establish a multi-objective optimization model for the service configuration problem, taking into account the timing constraints of the service process, the time-varying constraints of service resources, the configuration constraints of service components, the customer's service time and total budget constraints, etc.

[0127]

[0128]

[0129]

[0130]

[0131] x hij ≥δ iji′j′ x hi′j′ i, i′=1, 2, ..., N m ; j, j′ = 1, 2, ..., N i ;i&j≠i′&j′;h=1,2,...,H (5)

[0132] x hij ≤2-δ′ iji′j′ -x hi′j′ i, i′=1, 2, ..., N m ; j, j′ = 1, 2, ..., N i ;i&j≠ i′&j′;h=1,2,...,H (6)

[0133]

[0134]

[0135]

[0136]

[0137]

[0138] s h1 =0,s hi ≥0 i=2,...,N m +1; h=1, 2, ..., H (12)

[0139] x ij =0 or 1 i=1, 2, ..., N m ; j = 1, 2, ..., Ni ; h=1,2,...,H (3)

[0140] Specific parameter symbol meaning

[0141] x hij : Main decision variable, i = 1, 2, ..., N m ; j = 1, 2, ..., N i , h=1, 2, ..., H. If the jth module component of the i-th service module is selected in the configuration scheme, then x hij =1; otherwise x hij =0.

[0142] y hijt : The auxiliary variable represents the task end status of the jth module component of the i-th service module at time block t, that is, if time block t is the end time of the module component, then y hijt =1; otherwise y hijt =0.

[0143] In addition: N m : The number of required service modules; N i : Service module component, i=1, 2, ..., N m ;d ij : Service processing time of service module component, i=1, 2, ..., N m ; j = 1, 2, ..., N i ;c ij : Service unit cost of service module component, i = 1, 2, ..., N m ; j = 1, 2, ..., N i ; C fix : Fixed cost of service; H: Number of customers who need the service configured; C h : The upper limit of the total cost budget required by the customer, h = 1, 2, ..., H; T h : The upper limit of the total service time required by the customer, h = 1, 2, ..., H;

[0144] K1: The number of customer demands is; K2: The number of service attributes is; w k1 : The relative importance of customer needs k1 = 1, 2, ..., K1; q1 k1k2 : The correlation between customer needs and service attributes, k1=1,2,...,K1; k2=1,2,...,K2; q2 k2ij : The relationship between service attributes and service module components k2=1, 2, ..., K2; i=1, 2, ..., N m ; j = 1, 2, ..., N i; The temporal relationship of activities in the service process, Indicates that the i-th service module is the predecessor of the i′-th service module, otherwise i, i′ = 1, 2, ..., N m ;i≠i′;s hi : The start time of the service module i = 1, 2, ..., N m , h=1, 2, ..., H; T: maximum service period, integer; L: number of resource types, integer; R lt : the total amount of the lth resource at time t, l = 1, 2, ..., L; t = 1, 2, ..., T; r ijl : The resources required by the jth module component of the i-th service module during execution i=1, 2, ..., N m ; j = 1, 2, ..., N i ; l = 1, 2, ..., L; δ iji′j′ : Compatibility relationship of service module components, i, i′=1, 2, ..., N m ; j, j′ = 1, 2, ..., N i ;i&j≠i′&j′. When there is a compatibility relationship between module components (if the j′th module component of the i′th service module is selected in the configuration scheme, the jth module component of the i′th service module must also be selected), δ iji′j′ =1, otherwise δ iji′j′ =0;δ′ iji′j′ : Exclusive relationship of service module components, i, i′=1, 2, ..., N m ; j, j′ = 1, 2, ..., N i ; i&j≠i′&j′. When there is an exclusive relationship between module components (if the j′th module component of the i′th service module is selected in the configuration scheme, the jth module component of the i′th service module cannot be selected), δ′ iji′j′ =1; otherwise δ′ iji′j′ =0;

[0145] Step 4: Improve the established multi-objective service configuration optimization model into a linear programming model;

[0146] By analyzing the multi-objective optimization model (P), we know that constraint a4 is a nonlinear constraint, but the decision variables in this constraint are all 0-1 variables, which can be easily converted into equivalent linear constraints. Define a new auxiliary decision variable z hijt i=1,2,...,N m ; j = 1, 2, ..., N i; t=1,2,…,T;h=1,2,…,H, replace constraint a4 in the model with the following constraint:

[0147]

[0148]

[0149] M(2-x hij -y hijt )≥1-z hijt i=1,2,...,N m ; j = 1, 2, ..., N i ;t=1,2,...,T;h=1,2,...,H

[0150] (a4-2)

[0151] z hijt ≤x hij , z hijt ≤y hijt i=1,2,...,N m ; j = 1, 2, ..., N i ;t=1,2,...,T;h=1,2,...,H

[0152] (a4-3)

[0153] z hijt =0 or 1

[0154] (a4-4)

[0155] After the transformation, the decision variables of the multi-objective linear optimization model are 0-1 quantities, the coefficients can be normalized into integer type parameters, and all Pareto solutions can be obtained using an exact algorithm.

[0156] Step 5: QFD is a method that integrates quantitative and qualitative information, some of which is obtained by human experts based on their personal experience and has inherent uncertainty. Since there is no historical data for estimating the elements of HoQ1 and HoQ2, which represent a subjective uncertainty, it is best to represent them as fuzzy numbers and assume that the relationship between customer needs and service attributes in HoQ1 and HoQ2 is a triangular fuzzy number, where the most pessimistic values ​​are: Most likely value: Most optimistic value: Consider the optimization model based on fuzzy QFD, where the optimization objectives are obj3, obj4, and obj5; the influencing constraints are, where the detailed parameter information and model are as follows

[0157] (1) Model parameter information:

[0158] The relationship between customer needs and service attributes in HoQ1 is a triangular fuzzy number, expressed as:

[0159]

[0160] The relationship between customer needs and service attributes in HoQ2 is a triangular fuzzy number, expressed as:

[0161]

[0162] (2) Required optimization objectives

[0163] Obj3: The optimization model formula based on fuzzy HOQ is as follows:

[0164]

[0165] Among them I h It is not a triangular fuzzy number, but an LR type fuzzy number. h It can be expressed as

[0166] Most pessimistic value

[0167] Most likely value

[0168] Most optimistic value

[0169] Obj4: Based on Theorem 1 and Proposition 1, Model III is restated as follows:

[0170] Max I *

[0171] The impact constraints considered are:

[0172]

[0173]

[0174] (The rest of the constraints are the same as the basic model)

[0175] Obj5: The three-objective optimization model formula is as follows:

[0176]

[0177]

[0178]

[0179] The impact constraints considered are:

[0180]

[0181]

[0182]

[0183]

[0184] (The rest of the constraints are the same as the basic model)

[0185] in, All are continuous decision variables.

[0186] Step 6: Apply different algorithms to solve the configuration model based on the problem size. For small and medium-sized problems, use the space partitioning algorithm to find the exact Pareto solution set of the multi-objective model; for large-scale problems, design the NSGA-II algorithm to find the non-dominated solution set of the multi-objective model;

[0187] (1) For small and medium-sized service configuration problems, the present invention designs an objective function value space partitioning algorithm for optimization, converting the multi-objective optimization problem into a single-objective optimization problem. Each time the single-objective linear integer programming problem is solved, a new non-dominated solution is obtained. The search space is then partitioned using the new non-dominated solution, the space dominated by the non-dominated solution is eliminated, and the non-dominated solution set is updated. The above process is repeated until no new non-dominated solutions are generated, and the complete Pareto solution set of the multi-objective optimization problem is obtained. When solving the single-objective linear integer programming model of each sub-problem, optimization software such as CPLEX can be used for calculation.

[0188] (2) For large-scale service configuration problems, the present invention designs the NSGA-II algorithm for optimization. Based on NSGA, a fast non-dominated sort is performed. The congestion comparison operator and the elite retention strategy are introduced, which significantly reduces the computational complexity of the algorithm and improves its performance. The chromosomes used to solve the service configuration model in the present invention mainly adopt an integer coding structure, the model constraints are handled by the penalty function method, and the genetic operations use single-point crossover and neighborhood mutation.

[0189] Step 7: Solution set reduction and interactive selection. For the Pareto solution set or non-dominated solution set obtained, the K-means clustering method is first used to determine the cluster center. Then, the configuration solution closest to the example cluster center is found, reducing the solution set to a relatively small number of candidate solutions. The effectiveness of the two algorithms is demonstrated in numerical experiments and industrial examples. Finally, the customer adjusts the target and interactively selects the satisfactory solution.

[0190] Finally, the model of the present invention is explained through a practical case. According to the service feature analysis, the service modular design concept is used to divide the decoration service of a decoration company into 16 service modules, corresponding to 46 module components and their attribute values, as shown in Table 1. There is a time sequence relationship between the modules, as shown in Table 2. Each service module component requires different human resources. The model case of the present invention mainly considers the electrical resources required in the entire decoration process, and the components I in modules (4) and (15) are 41 , I 42 and I 152 Set the required human resources to 2, 2 and 3 respectively.

[0191] In order to facilitate modeling, the optional modules and mandatory modules are generalized, that is, the optional modules are assumed to be mandatory, and a module component is added to the optional modules as "not selected", and its attribute value is 0. Among the module components, the decoration module components with compatible relationships are: 42 and I 142 , I 13 and I 142 , I 62 and I 103 , I 94 and I 113 , I 11 and I 153 , I 13 and I 73 , I 11 and I 132 ; The decoration module components that have an exclusive relationship are: I 93 and I 112 , I 11 and I 83 , I 11 and I 162 .

[0192] Table 1. Decoration service module and its module component information

[0193]

[0194]

[0195]

[0196] Note: Time unit: day; Cost unit: RMB 100; Service modules marked with * are optional modules

[0197] Table 2. Adjacency matrix of relationships between service modules

[0198] <![CDATA[N m ]]> 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 - 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 - 1 1 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 - 0 1 0 0 0 0 0 0 0 0 0 0 0 4 0 0 0 - 1 0 0 0 0 0 0 0 0 0 0 0 5 0 0 0 0 - 1 1 0 0 0 0 0 0 0 0 0 6 0 0 0 0 0 - 0 1 0 0 0 0 0 0 0 0 7 0 0 0 0 0 0 - 0 1 0 0 0 0 0 0 0 8 0 0 0 0 0 0 0 - 1 0 0 0 0 0 0 0 9 0 0 0 0 0 0 0 0 - 1 1 0 0 0 0 0 10 0 0 0 0 0 0 0 0 0 - 0 0 0 1 0 0 11 0 0 0 0 0 0 0 0 0 0 - 1 0 0 0 0 12 0 0 0 0 0 0 0 0 0 0 0 - 1 0 0 0 13 0 0 0 0 0 0 0 0 0 0 0 0 - 0 1 0 14 0 0 0 0 0 0 0 0 0 0 0 0 0 - 1 0 15 0 0 0 0 0 0 0 0 0 0 0 0 0 0 - 1 16 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 -

[0199] Through step 2, the two decoration customers scored the six requirements as Sc1 = [10, 8, 9, 10, 7, 7] and Sc2 = [6, 7, 6, 6, 10, 10]. The performance values ​​of the 46 components in the decoration module were calculated by the two-level QFD quality house formula as shown in Table 3, where the quality house parameter q1 k1k2 and q2 k2ij The value of is obtained by comprehensive analysis of relevant experts convened by the decoration company. Among them, are the most pessimistic, most likely and most optimistic performance values, respectively.

[0200] Table 3 Performance values ​​of decoration service module and its module components

[0201]

[0202]

[0203]

[0204]

[0205] Through steps 3 and 4, a multi-objective optimization model for decoration service configuration based on fuzzy QFD is established. The three objectives are based on the constraints a9, a10, a11 and The values ​​of are calculated and the data in the previous steps are used as input parameters of the optimization model.

[0206] Through step 5, the three-objective model based on fuzzy QFD is solved. This situation can be addressed using the proposed spatial partitioning algorithm. The algorithm was programmed using the Pareto set and Matlab™. It took 2776.8 seconds to obtain accurate data, as shown in Table 4. The exhaustive search algorithm also yielded the same results, verifying its accuracy.

[0207] Table 4 Algorithm results

[0208]

[0209] The optimization results show that there are two non-optimal solutions, which have little impact on the configuration of the second customer and the two objectives ( and Since there are only two non-dominated solutions, it is easy for the decision maker to make a choice. The detailed information of the corresponding solutions, including the selected module instance, the earliest start time (EST), service cost, and resource consumption, are shown in Table 5.

[0210] Table 5. Details of Pareto setting

[0211]

[0212]

[0213] Based on the above situation, a sensitivity analysis experiment was conducted on the budget and the required service period. In order to simplify the experiment, C h1 and T h1 , C h2 Change in units of 5K yuan, T h2 Changes are made in 2-day increments.

[0214] Based on the data in Table 1, the minimum and maximum costs for all service plans are calculated as follows: and Therefore, in this time interval 26 experimental cases were divided into two groups, and the budget sensitivity analysis results are as follows: Figure 5 As shown in Table C1 (see Appendix C), it can be seen that budgeting has a positive impact on CSI, that is, the larger the budget, the higher the CSI the company can provide. h2 ≥145K, the experimental results are constantly changing, indicating that C h2 =145K can be set as the customer's maximum budget, and any increase in the budget will not have an impact on the overall CSI. z-za Value and It is very small (≤0.0035), which means that the budget is not sensitive in the [85K110K] budget range. On the other hand, because there is IGD in the [110K, 115K] range z-za The maximum value is therefore the most sensitive to budget.

[0215] Similarly, according to the data in Table 1, the minimum and maximum service durations of all service solutions are and Therefore, 17 experimental cases were divided into two intervals, and the results of the budget sensitivity analysis are shown in Figure 6 and Table C2 (see Appendix C). It can be seen that the required service period also has a positive impact on CSI, that is, the longer the required service period, the higher the CSI the company can provide for renovation plans. h2 = 44, there is no feasible decoration scheme to recommend. h2 ≥62, the experimental results keep changing, indicating that T h2 =62 can be set as the maximum service duration for this customer, and the increase in service duration has no effect on CSI. z-za In Table C2, the most sensitive point is T h2= 56, i.e., when the service duration is T h2 =56 is extended by 2 days, and the CSI can be greatly improved.

[0216] Through steps 6 and 7, NSGA-II is used to solve large-scale problems. Based on the fundamental characteristics of genetic algorithms, the population size N, generation size G, crossover probability px, and mutation probability pm significantly influence the performance of NSGA-II. Taguchi-designed experiments are used to adjust the combination of these four parameters to achieve optimal performance, as evaluated by run time and generation distance (GD). A smaller GD indicates a closer approach to the exact solution. Five different level values ​​are set for each problem, based on the problem size and the characteristics of NSGA-II, as shown in Table 6.

[0217] Table 6. Parameter levels

[0218]

[0219] In Table 6, the four parameters G, N, px, and pm are used as control factors, and the running time (time) and GD are used as two noise factors. An orthogonal table of L25 (54) is formed, and each parameter combination is run three times. The results are shown in Table 7.

[0220] Table 7. Experimental results of L25(54) parameter combinations

[0221]

[0222]

[0223] The analysis results are as follows:

[0224] 1) The average GD value mainly affects the diagram, such as Figure 7 When px = 0.75 and pm = 0.15, the GD value is small and relatively stable.

[0225] 2) The main influence diagram of the average time value is as follows Figure 7 As shown in (b), the larger the values ​​of N and G, the longer the execution time. When N is in the range of 500-600 and G is in the range of 400-600, the increase in algorithm execution time is more significant.

[0226] 3) Through integration Figure 7 In the two figures, we set the values ​​of the four parameters to N = 500, G = 200, px = 0.75, pm = 0.15 to achieve better efficiency and effect of NSGA-II.

[0227] Based on the selected parameter settings (N=500, G=200, px=0.75, pm=0.15), NSGA-II was executed for five rounds with an average GD of 0.0197 and an average runtime of 100.6 seconds. For CSI, the differences between the most pessimistic, most likely, and most optimistic values ​​and the exact Pareto front solution are only between 0.5% and 1.5%. Therefore, for the industry cases in this section, NSGA-II can not only obtain service recommendations within a reasonable runtime, but also the gap between the obtained service recommendations and the optimal service solutions is very small. NSGA-II helps companies cope with more module instances and customer combinations.

[0228] To further evaluate the efficiency of the proposed exact and metaheuristic algorithms, service configuration simulation cases were constructed with constraints on compatibility and exclusivity between module instances, cost constraints, timing constraints, resource constraints, and multi-client resource constraints. Timing constraints were randomly generated. Small-scale cases included M = 10 and M = 40, while large-scale cases included M = 30 and M = 40. The control parameters of NSGA-II were (px = 0.75, pm = 0.15, N = 300, G = 200) for the small-scale case and (px = 0.75, pm = 0.15, N = 600, G = 400) for the large-scale case. For each case, the number of clients was set to H = 2 and H = 3, generating eight small-scale cases and eight large-scale cases for the experiment.

[0229] Table 8 shows the results of randomly generated small-scale cases. For the case where (M = 15, N_I = 40, H = 3) is used, running the spatial partitioning algorithm takes more than 30 days and the corresponding exact Pareto set is unknown. Compared with NSGA-II, the spatial partitioning algorithm has a great advantage for some small-scale problems. For example, the running time of the spatial partitioning algorithm is only 0.5 seconds, while for the unavailable case (M = 10, N_I = 40, H = 3), the running time is only 1.5 seconds. I =24,H=2 and M=10,N I =25, H = 2). However, as the number of module instances or the number of clients increases, the efficiency of the space partitioning algorithm becomes very low. For example, for (M = 10, N I =24, H=3) and M=10, N I =25, H =3), the runtimes were 712.2 seconds and 10,047 seconds, respectively, significantly longer than NSGA-II. Furthermore, due to the small number of solutions in the Pareto set, the algorithm's performance was analyzed by calculating the GD, which was found to be very low. This indicates that the average minimum distance between the solutions obtained by NSGA-II and the exact solution is very close. Therefore, NSGA-II is able to obtain a near-optimal solution within a reasonable runtime.

[0230] Table 8. Results of randomly generated small-scale cases

[0231]

[0232]

[0233] *M--module quantity; N_I--total number of module instances; N_MIC--number of module instance combinations;

[0234] H is the number of customers; |NP| is the number of non-dominated points; |NA| is the number of approximate non-dominated points

[0235] Table 9 presents the results for a randomly generated large-scale case. The number of solutions in the Pareto set is relatively small; therefore, HV and HVD are used to estimate the convergence of the NSGA-II algorithm. The reference point for calculating HV is set to (1,1,1). Comparing the smaller values ​​of HVD between the calculated solution (N = 600, G = 400) and the reference solution (N = 1200, G = 800), the closer the calculated results are to the Pareto set, indicating that the improved NSGA-II algorithm has good convergence.

[0236] Table 9. Results of randomly generated large-scale cases

[0237]

[0238]

[0239] Through the above configuration optimization process, the optional solutions for decoration services are enriched, the flexibility of the decoration process is improved, and rapid automatic configuration according to customer needs can be achieved. The multi-customer service configuration optimization method considering service process information and time-varying service resource constraints generates customer-satisfied decoration service configuration solutions. At the same time, it also provides decoration companies with more time-varying resource scheduling solutions, which is conducive to improving their economic benefits.

Claims

1. A modular service configuration optimization method considering multi-user and fuzzy quality house information, characterized in that: The steps include: Step 1: Establish a service modular structure that uses the design structure matrix method to represent the complex service process, and define the service components available in the service module and their basic attribute values; Step 2: Construct a two-level QFD quality house to map the importance of different customer needs to the service component preferences of each service module; Step 3: Establish a multi-customer service configuration optimization model that comprehensively considers the time constraints of the service process, the configuration constraints of service components, the budget constraints, and the service duration constraints; Step 4: Improve the established multi-objective service configuration optimization model into a linear programming model; Step 5: Based on the consideration of the fuzzy quality house information, the model is expanded and further transformed into a three-objective model; Step 6: Construct a set of recommended solutions for different algorithms to calculate service configuration optimization problems of different scales; Step 7: Construct a K-means clustering method to reduce the solution set to form a candidate solution set. Finally, the customer adjusts the target and interactively selects the satisfactory solution. In step 3, in the service configuration optimization model considering service process information: (1) Model parameter information is as follows: N m : The number of required service modules; N i :Service module component, i=1,2,…,N m ; d ij : Service processing time of service module component, i=1,2,…,N m ; j=1,2,…,N i ; c ij : Service unit cost of service module component, i=1,2,…,N m ; j=1,2,…,N i ; C fix : Fixed fee for services; H: the number of customers who need the configured service; C h : The upper limit of the total cost budget required by the customer, h = 1, 2, ..., H; T h : The upper limit of the total service time required by the customer, h = 1, 2, ..., H; K1: the number of customer demands; K2: the number of service attributes; w hk1 : The relative importance of customer needs q1 k1k2 : The correlation between customer needs and service attributes, k1=1,2,…,K1; k2=1,2,…,K2; q2 k2ij :The relationship between service attributes and service module components k2=1,2,…,K2;i=1,2,…,N m ; j=1,2,…,N i ; The temporal relationship of activities in the service process, Indicates that the i-th service module is the predecessor of the i′-th service module, otherwise s hi :Start time of service module i=1,2,…,N m , h=1,2,…,H; T: maximum service period, integer; L: the number of resource types, integer; P lt : the total amount of the lth resource at time t, l = 1, 2, ..., L; t = 1, 2, ..., T; r ijl : The resources required by the jth module component of the i-th service module during execution i=1,2,…,N m ; j=1,2,…,N i ; l=1,2,…,L; δ iji′j′ : Compatibility relationship of service module components, i,i′=1,2,…,N m ; j, j′=1,2,…,N i ; i&j≠i′&j′; When there is a compatible relationship between module components, that is, if the j′th module component of the i′th service module is selected in the configuration scheme, then the jth module component of the i′th service module must also be selected, δ iji′j′ =1, otherwise δ iji′j′ =0; δ′ iji′j′ : Exclusive relationship between service module components, i,i′=1,2,…,N m ; j, j′=1,2,…,N i ; i&j≠i′&j′; When there is an exclusion relationship between module components, that is, if the j′th module component of the i′th service module is selected in the configuration scheme, the j′th module component of the i′th service module cannot be selected, δ′ iji′j′ =1; otherwise δ′ iji′j′ =0; x hij :Main decision variable, i=1,2,…,N m ; j=1,2,…,N i ,h=1,2,…,H;If the jth module component of the ith service module is selected in the configuration scheme, then x hij =1; otherwise x hij =0; y hijt : The auxiliary variable represents the task end status of the jth module component of the i-th service module at time block t, that is, if time block t is the end time of the module component, then y hijt =1; otherwise y hijt =0; (2) The multiple targets to be optimized include obj1 and obj2, where: Optimization objective obj1: Maximize the minimum value of the customer satisfaction index (CSI) of each customer, expressed as the formula: Optimization objective obj2: maximize R, expressed as formula: (3) The impact constraints considered are a1, a2, a3, a4, a5, a6, a7, and a8, where: Impact constraint a1: Only one module component in each service module can be selected. This relationship is expressed as: Impact constraint a2: The temporal relationship of service module nodes in the service process. This relationship is expressed as: Impact constraint a3: The compatible and exclusive configuration relationship between service module components. This relationship is expressed as: x hij ≥δ iji′j′ x hi′j′ i,i′=1,2,…,N m ;j,j′=1,2,…,N i ;i&j≠i ′ &j′;h=1,2,…,H x hij ≤2-δ′ iji′j′ -x hi′j′ i,i′=1,2,…,N m ;j,j′=1,2,…,N i ;i&j≠i′&j′;h=1,2,…,H Impact constraint a4: During the entire service process, the resources occupied by each module component cannot exceed its upper limit. This relationship is expressed as: Influence constraint a5: limits the auxiliary variable y ijt The value of conforms to its definition, and this relationship is expressed as: Impact constraint a6: The total cost and total duration of the service cannot exceed the upper limit given by the customer. This relationship is expressed as: Impact constraint a7: Initial setting of service component start time and non-negative constraint. This relationship is expressed as: s h1 =0,s hi ≥0i=2,…,N m +1;h=1,2,…,H Influence constraint a8: gives the specific definition of configuration decision variables and restricts decision variables. This relationship is expressed as: x ij =0or 1i=1,2,…,N m ;j=1,2,…,N i ;h=1,2,…,H。 2. The method according to claim 1, characterized in that In step 2, the importance of different needs to customers is mapped to the service component preferences of each service module; First, establish a service planning quality house, where the "left wall" represents customer needs and the "ceiling" represents basic service attributes. Then, establish a module configuration quality house, where the "left wall" represents basic service attributes and the "ceiling" represents the service components of each service module. Through this two-level service quality house, the importance of customer needs can be mapped to the service component preferences of the service module. By drawing on the first two levels of the four-level quality house structure of QFD product quality design, we successively build the service planning quality house and the module configuration quality house, mapping the importance of customer needs to the service components of the service module. The specific steps are as follows: (1) First, analyze and clarify customer needs; different customers score the importance of all needs based on their own needs, and normalize the importance of different needs; (2) Secondly, the mapping relationship between the two quality houses is identified and evaluated by decoration experts, and the values ​​of the quality house parameters are obtained by comprehensive analysis by relevant experts convened by the enterprise; (3) Finally, the corresponding performance value is calculated based on the module configuration quality house.

3. The method according to claim 2, characterized in that Step 4 is as follows: By analyzing the multi-objective optimization model, we know that constraint a4 is a nonlinear constraint, but the decision variables in this constraint are all 0-1 variables, which can be easily converted into equivalent linear constraints. Define a new auxiliary decision variable z hijt i=1,2,…,N m ; j=1,2,…,N i ; t=1,2,…,T;h=1,2,…,H, replace constraint a4 in the model with the following constraint: M(2-x hij -y hijt )≥1-z hijt i=1,2,…,N m ;j=1,2,…,N i ;t=1,2,…,T;h=1,2,…,H (a4-2) from hijt ≤x hij ,from hijt ≤y hijt i=1,2,…,N m ;j=1,2,…,N i ;t=1,2,…,T;h=1,2,…,H(a4-3) from hijt =0 or 1(a4-4) The decision variables of the converted multi-objective linear optimization model are 0-1 quantities.

4. The method according to claim 3, characterized in that In step 5, QFD is a method that integrates quantitative and qualitative information, some of which is obtained by human experts based on personal experience and has inherent uncertainty. The elements of HoQ1 and HoQ2 do not have historical data for estimation and are subjective uncertainties. Therefore, they are represented as fuzzy numbers and it is assumed that the relationship between customer needs and service attributes in HoQ1 and HoQ2 is a triangular fuzzy number, where the most pessimistic values ​​are: Most likely value: Most optimistic value: Consider the optimization model based on fuzzy QFD, where the optimization objectives are obj3, obj4, and obj5; the influencing constraints are, where the detailed parameter information and model are as follows (1) Model parameter information: The relationship between customer needs and service attributes in HoQ1 is a triangular fuzzy number, expressed as: Where k1 = 1, 2, ..., K1; k2 = 1, 2, ..., K2 The relationship between customer needs and service attributes in HoQ2 is a triangular fuzzy number, expressed as: Where, k2=1,2,…,K2; i=1,2,…,N m ; j=1,2,…,N i (2) Required optimization objectives Obj3: The optimization model formula based on fuzzy HOQ is as follows: Among them, I h is a triangular fuzzy number or LR type fuzzy number; I h It can be expressed as in: Most pessimistic value Most likely value Most optimistic value Object 4: Restate Model III as follows: Max I * The impact constraints considered are: The remaining constraints are the same as the basic model; Obj5: The three-objective optimization model formula is as follows: The impact constraints considered are: The remaining constraints are the same as the basic model; in, All are continuous decision variables.

Images