Cloud manufacturing service combination and scheduling integrated optimization method

Abstract

The invention relates to the technical field of cloud manufacturing, in particular to a cloud manufacturing service combination and scheduling integrated optimization method. Comprising the steps that 1, a complex matching network of cloud manufacturing services and cloud manufacturing tasks is established, the cloud manufacturing services are divided into cloud manufacturing production services and cloud manufacturing logistics services, and the cloud manufacturing tasks also comprise cloud manufacturing production tasks and cloud manufacturing logistics tasks; 2, a cloud manufacturing service combination and scheduling integration optimization model is established, QoS of all cloud manufacturing services in the network is quantified through weighted summation of a service quality parameter and a service reliability parameter, the two parameters are dynamically described by a time domain, and an initial value is an evaluation value obtained in advance; and step 3, solving the integrated optimization model by adopting a multi-population co-evolution algorithm based on meta-knowledge adaptive migration to obtain a multi-objective optimization Pareto solution set, and decoding the multi-objective optimization Pareto solution set into an optimization scheduling scheme. And performing combined configuration and scheduling optimization on services in the cloud manufacturing platform.

Description

Technical Field

[0001] The present invention relates to the field of cloud manufacturing technology, and in particular to a cloud manufacturing service combination and scheduling integrated optimization method. Background Art

[0002] Driven by the external market's demand for complex and personalized manufacturing and enabled by the development of emerging technologies, the cloud manufacturing service model has been proposed and widely used. Cloud manufacturing integrates manufacturing resources from multiple different fields and regions to achieve centralized management and optimal allocation of distributed manufacturing resources, gradually breaking through the limitations of resource sharing and utilization efficiency in the manufacturing industry, thereby promoting comprehensive resource sharing and improving resource utilization. Cloud manufacturing service combination and scheduling are the core links in the operation of the cloud manufacturing service model.

[0003] Cloud manufacturing service composition and scheduling are two closely connected processes, which affect each other and directly affect the quality and efficiency of the final cloud manufacturing service solution. However, existing methods usually carry out manufacturing service composition and cloud manufacturing service scheduling in two stages, which limits the quality of the service solution obtained by the optimization decision. At the same time, industrial robots have been widely used in cloud manufacturing services as cloud manufacturing resources. Multiple degrees of freedom make industrial robots have high manufacturing flexibility and can flexibly complete different types of manufacturing tasks according to manufacturing needs. At the same time, the long-term processing and production of process robots with complex mechanical structures will lead to fluctuations in their service quality. However, the current cloud manufacturing service scheduling method assumes that the cloud manufacturing service quality is stable and unchanged, and does not consider the time-domain fluctuation characteristics of service quality caused by cloud manufacturing resource loss and degradation, which makes the actual cloud manufacturing service quality deviate from the expected cloud manufacturing service quality. On the other hand, the integrated optimization of cloud manufacturing service composition and scheduling is a complex optimization problem. Continuing to use the existing intelligent optimization algorithm to solve it will often fall into the local optimum, and it is difficult to obtain the optimal result for this problem. Summary of the invention

[0004] Purpose of the Invention The present invention provides a new method for integrated optimization of cloud manufacturing service combination and scheduling, which combines and configures services in a cloud manufacturing platform and optimizes scheduling.

[0005] The cloud manufacturing service combination and scheduling integrated optimization method of the present invention includes the following:

[0006] Step 1: Establish a complex matching network of cloud manufacturing services and cloud manufacturing tasks, where cloud manufacturing services are divided into cloud manufacturing production services and cloud manufacturing logistics services, and cloud manufacturing tasks also include cloud manufacturing production tasks and cloud manufacturing logistics tasks;

[0007] Step 2: Based on the complex matching network, with the optimization objectives of maximizing overall QoS, minimizing cost and minimizing delivery time difference, a cloud manufacturing service combination and scheduling integrated optimization model is established, wherein the QoS of each cloud manufacturing service in the network is quantified by the weighted sum of service quality parameters and service reliability parameters, both parameters are described dynamically in the time domain, and the initial value is the evaluation value obtained in advance;

[0008] Step 3: Use a multi-population co-evolutionary algorithm based on meta-knowledge adaptive migration to solve the integrated optimization model, obtain a multi-objective optimization Pareto solution set, and decode it into an optimal scheduling solution.

[0009] Furthermore, the complex matching network includes multiple production resources, each of which corresponds to a production service, and the parameters describing the production resources include: production resource number PR id , Provider PR P , Function type PR FU , position coordinates [PR x ,PR y ]、Production Capacity PR t ; Parameters describing the production service include, production service number PS id , Reliability PS Re , Quality PS Qu , Cost PS C ;

[0010] It also includes multiple logistics service providers and logistics services. The parameters describing the logistics service providers include the label LR id , Logistics service range radius LR RA , logistics service node set LR TP and logistics transfer speed LS S ; Parameters describing logistics services include, label PS id , Cost LS C and service mode LS SP ;

[0011] Furthermore, reliability PS Re and quality PS Qu The time domain dynamic description is as follows:

[0012] Let the initial service quality of the new equipment be Q0, and the service quality degradation curve is described by the Gamma process. The shape parameter α and size parameter β are fitted by the manufacturing resource curve data, and the Gamma process is used to describe the degradation behavior of the manufacturing resource service quality, as shown in the following formula:

[0013]

[0014] where Γ(·) is the Gamma function;

[0015] The real-time dynamic cloud manufacturing resource service quality is expressed by the equipment status index for service quality, that is, the initial service quality minus the degradation amount, as shown below.

[0016]

[0017] Similarly, with the initial reliability of the new equipment being R0 and the reliability degradation curve described by the Gamma process, the equipment status index for reliability can be obtained to represent the real-time dynamic reliability of the equipment. When the manufacturing service reliability decreases to a certain extent, it also needs to be maintained and replaced.

[0018] When the equipment status index X l (t) drops to the equipment maintenance point M p , the equipment is maintained. The status index after equipment maintenance is Z l <QoS0, where QoS0 = Q0 or R0, representing the initial quantity; the improvement factor is used to describe the improvement of the status index after equipment maintenance. The calculation formula for the improvement factor is (X l (t) - Z l ) / X l (t), X l (t) is the equipment status index before maintenance. When the status index after maintenance drops to QT u , it needs to be updated, and the status index after update is restored to QoS0.

[0019] Furthermore, the optimization objective is expressed as:

[0020]

[0021] where TCQoS is the overall network QoS, calculated as follows:

[0022] QoS = k1PS Qu + k2PS Re

[0023] k1 + k2 = 1

[0024] TCQoS = ∑QoS i

[0025] where PS Qu , PS Re both have a value range of [0.5, 1], and k1, k2 are weight coefficients;

[0026] TC is the cost of all cloud manufacturing tasks. The cost of a single service C = PST c + LST c, where PST c The cost of production services, LST c is the logistics service cost, including:

[0027] PS C =C Fle *C Mul *PS Re *PS Qu *C fun *Workload

[0028]

[0029] Among them, C fun is the unit time cost corresponding to each service type, C Fle The coefficient corresponding to the service flexibility type, C Mul Indicates the coefficient corresponding to the service function type, and Workload is the workload;

[0030] as well as,

[0031] LS C =Cc*LS S *Dis,

[0032] Where Cc is the conversion factor between transit speed and transit cost, and Dis is the distance;

[0033] DTD is the delivery time difference, which is:

[0034]

[0035] Where I is the number of cloud manufacturing tasks, The delivery time of task i is generated by the cloud, The actual completion time of cloud manufacturing task i is:

[0036] T CT =T Pro +T Log +T Idl ,

[0037] Where T Pro is the cloud manufacturing production task time, T Log Time for cloud manufacturing logistics service, T Idl The waiting time between task executions.

[0038] Furthermore, the constraints of the optimization model include:

[0039] They respectively indicate that only one production / logistics subtask can be assigned to one production / logistics service at a time, i is the cloud manufacturing task index, j is the cloud manufacturing subtask index, m is the cloud manufacturing production service index, and n p is the total number of cloud manufacturing production services, k is the cloud manufacturing logistics service index, N l The total number of cloud manufacturing logistics services,

[0040] PZ ijm =1 indicates cloud manufacturing production subtask ST ij is assigned to the mth production service, otherwise it is 0,

[0041] LZ ijk =1 indicates cloud manufacturing logistics subtask LZ ij is assigned to the kth logistics service, otherwise it is 0;

[0042] Indicates that only one production service can be assigned to process a cloud manufacturing task at the same time, N t represents the number of cloud manufacturing tasks, represents the number of subtasks within the i-th cloud manufacturing task;

[0043] max{PE i′j′m ,LE ijk}≤PS ijm , m=k means that the task start time needs to be after the completion of other tasks and the preceding logistics tasks; PS ijm Perform subtask ST for the mth production service ij Start time, PE i′j′m Perform subtask ST for the mth production service i′j′ End time, LE ijk For the k-th logistics service, perform the subtask LE ij End time of

[0044] PE ijk ≤LS ijm , k = m-1, indicating that the logistics task needs to be started after the immediate production task is completed;

[0045] Constraint LZ ijk With PZ ijk is a binary decision variable.

[0046] Furthermore, step 3 includes:

[0047] Step 3.1: Individual coding;

[0048] The genetic coding content of a single entity includes the production task allocation code, the logistics task allocation code and the production task relative priority code, all of which are represented by real numbers between [0,1]. The real number codes of the production and logistics task allocation parts are used to select services from the candidate service set to perform cloud manufacturing subtasks;

[0049] The relative priority code of the production task is used to determine the scheduling priority of the cloud manufacturing production task, and is also represented by a real number between [0,1];

[0050] The corresponding decoding steps include:

[0051] Determine the number of candidate manufacturing service sets according to the manufacturing service function requirements, multiply the number of candidate service sets by the corresponding code to obtain a selected cloud manufacturing service, and reorder the production services according to their relative priorities to generate a scheduling sequence;

[0052] Determine the number of candidate logistics services based on the coordinate information of adjacent production services, also obtained based on the coding values;

[0053] According to the selected cloud manufacturing service capability information, cloud manufacturing tasks are inserted into the service provider's local production scheduling plan, a scheduling plan is generated, and the corresponding service quality, service cost, and delivery time difference are calculated;

[0054] Step 3.2: Initialize the population and divide it into S n subpopulation;

[0055] Step 3.3: Each sub-population selects different algorithms according to probability and iterates them a certain number of times;

[0056] Step 3.4: Meta-knowledge adaptive transfer and population update

[0057] Calculate the source population SP So With the target population SP Ta Distribution Center: In the formula is the value of the center of subpopulation k in dimension j; is the encoding value of the i-th individual of subpopulation k in dimension j; NP is the number of subpopulations;

[0058] Calculate the distance of each dimension from the individual to the center of the subpopulation to which it belongs: is the distance from the encoding value of the i-th individual in the k-th subpopulation in the j-dimension to the subpopulation center, is the encoding value of the i-th individual in the k-th subpopulation on the j-dimension;

[0059] Step 3.4.1 Meta-knowledge Adaptive Transfer

[0060] Step 3.4.1.1 traverse each sub-population and calculate the sub-population meta-knowledge probability distribution model;

[0061] Step 3.4.3.2 traverses each subpopulation, calculates the cross entropy between the meta-knowledge probability models of each subpopulation, and obtains the subpopulation cross entropy matrix; the cross entropy calculation formula between two subpopulations is as follows:

[0062]

[0063] P(x i ): multivariate Gaussian distribution function of population 1,

[0064] Q(x i ): multivariate Gaussian distribution function of population 2;

[0065] Step 3.4.3.3 Calculate the meta-knowledge adaptive migration threshold point based on the mean of all cross entropies in the cross entropy matrix. If the cross entropy between sub-populations is greater than the threshold point, the sub-populations will perform meta-knowledge migration between each other. The migration process generates a new individual InD according to the following formula: new :

[0066]

[0067] Where c Ta is the meta-knowledge of the source population, i.e., an i*j real number matrix, i is the number of individuals, j is the number of individual encoding dimensions, and mk is the population center in the target population, i.e., a j-dimensional real number matrix;

[0068] Step 3.5 evaluates the new individuals generated by the adaptive migration of meta-knowledge and updates the individuals in the sub-population to keep the population size unchanged;

[0069] The new individuals obtained by adaptive transfer of meta-knowledge are evaluated, and the best proportion of them are exchanged with the proportion of the worst fitness value in the original population.

[0070] Step 3.6: Determine the iteration stop condition. If it is met, stop the iteration. If not, return to step 3.3 for the next round of iterative evolution.

[0071] Furthermore, in step 3.2, the Scramble randomized Halton algorithm and the random generation method are used to initialize the population.

[0072] Furthermore, in step 3.2, the K-means clustering algorithm is used to divide the initialized population into sub-populations.

[0073] Furthermore, in step 3.3, each subpopulation adopts NSGA-Ⅲ or MOEA-DE for iterative evolution, and the probability of selecting the two algorithms is 50%.

[0074] Furthermore, in step 3.3, the number of iterations and evolutions is 15 times.

[0075] Furthermore, in step 3.4.1.1, the probability distribution model of the individual coding values ​​in the subpopulation is fitted using the multivariate normal distribution:

[0076]

[0077] Where N is the number of individuals in the subpopulation, x i,j is the value of the i-th individual in the subpopulation on the j-th dimension, u j is the mean of the encoding values ​​of all individuals in the subpopulation in dimension j, is the standard deviation of the coded values ​​of all individuals in the subpopulation in the dimension; ∑ is the covariance matrix of the subpopulation; μ = (μ1, μ2,…, μl).

[0078] Furthermore, the meta-knowledge adaptive migration threshold point is taken as λ < 1 times the mean of the sub-population cross entropy matrix.

[0079] Furthermore, the iteration stopping condition is that the number of iterations reaches a specified number.

[0080] Aiming at the temporal fluctuation characteristics of cloud manufacturing service quality based on industrial robots, this paper proposes a manufacturing service quality description model to more accurately quantify the quality of industrial robot cloud manufacturing services. Furthermore, a cloud manufacturing service combination and scheduling integrated optimization model is established. Compared with the two-stage solution method of cloud manufacturing service combination and scheduling, the service solution quality and service efficiency of the optimized solution can be improved, and the deviation between the actual service quality and the expected service quality caused by the fluctuation characteristics of cloud manufacturing services can be avoided.

[0081] The present invention also proposes a multi-population co-evolutionary algorithm for solving the cloud manufacturing service combinatorial scheduling integrated optimization problem. The algorithm adopts an innovative meta-knowledge adaptive migration mechanism, which can effectively solve the manufacturing service combinatorial scheduling optimization problem and avoid the population from falling into local optimality. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 Schematic diagram of the time domain change curve of cloud manufacturing production resource attributes in the embodiment

[0083] Figure 2 4 is a flow chart of a cloud manufacturing service in an embodiment.

[0084] Figure 3 The process of the meta-knowledge adaptive migration multi-population co-evolution algorithm in the embodiment

[0085] Figure 4 Schematic diagram of subpopulation meta-knowledge migration in the embodiment

[0086] Figure 5 It is a directed graph of cloud manufacturing tasks of Test Example 2 in the embodiment.

[0087] Figure 6 This is a network diagram for matching cloud manufacturing production resources with cloud manufacturing subtasks in Test Example 2 of the embodiment.

[0088] Figure 7 It is a schematic diagram of the solution set obtained by solving each algorithm in Test Example 2 in the embodiment.

[0089] Figure 8 It is a bar chart showing the proportion of Pareto solutions in the solution set obtained by each algorithm in Test Example 2 in the embodiment. DETAILED DESCRIPTION

[0090] The cloud manufacturing service composition and scheduling integrated optimization method in this example first establishes a complex matching network of cloud manufacturing services and cloud manufacturing tasks, then establishes an integrated optimization model for cloud manufacturing service composition and scheduling, and finally, a multi-population co-evolutionary algorithm based on meta-knowledge adaptive migration is used to solve the integrated optimization model to obtain a multi-objective optimization Pareto solution set.

[0091] The specific steps of this method are as follows:

[0092] Step 1: Establish a complex network relationship between cloud manufacturing services and cloud manufacturing tasks;

[0093] Specifically, cloud manufacturing services are divided into cloud manufacturing production services and cloud manufacturing logistics services, and cloud manufacturing tasks also include cloud manufacturing production tasks and cloud manufacturing logistics tasks;

[0094] Step 1.1: Establish a cloud manufacturing production service network model. The model components are shown in Table 1:

[0095] Table 1 Constituent elements of cloud manufacturing service network

[0096]

[0097]

[0098] Each resource and service has a one-to-one mapping relationship. When cloud manufacturing production resources are mapped to cloud manufacturing production services, the relevant attributes of the production resources are inherited to the cloud manufacturing services, and the cloud manufacturing services themselves also include relevant attributes such as QoS. The manufacturing service reliability and service quality values ​​are scored by domain experts, with a full score of 1 and a minimum score of 0.5. The cloud manufacturing production service cost calculation formula is as follows:

[0099] PS C =C Fle *C Mul *PS Re *PS Qu *C fun *Workload

[0100]

[0101] Where C fun is the unit time cost corresponding to each service type, in yuan / hour, C Fle The coefficient corresponding to the service flexibility type, C Mul Indicates the coefficient corresponding to the service function type, and Workload is the workload.

[0102] Step 1.1.1, establish a time domain dynamic description model of QoS (Quality of Service) such as cloud manufacturing production service quality and reliability.

[0103] The initial service quality of a new device is Q0, and the service quality degradation curve is described by the Gamma process. Whenever the service quality decreases to the maintenance point QT as the device is used, m When the equipment service quality after maintenance is lower than the equipment update point QT u After that, the equipment is updated and the service quality is restored to the initial value. The time domain change curve is as follows Figure 1 shown.

[0104] Specifically, the shape parameter α and size parameter β are fitted by the manufacturing resource curve data, and the Gamma process is used to describe the degradation behavior of the manufacturing resource service quality, as shown in the following formula:

[0105]

[0106] Where Γ(·) is the Gamma function.

[0107] Furthermore, the real-time dynamic cloud manufacturing resource service quality is expressed by the equipment status index for service quality, that is, the initial service quality minus the degradation amount (the cumulative distribution function of the Gamma process), as shown below:

[0108]

[0109] When the manufacturing service quality decreases to a certain level, it needs to be maintained to improve the service quality.

[0110] Similarly, taking the initial reliability of the new device as R0 and the reliability degradation curve described by the Gamma process, the device state index for reliability can be obtained to represent the real-time dynamic reliability of the device. When the manufacturing service reliability decreases to a certain extent, it also needs to be maintained to improve the service reliability.

[0111] Generally, when the device state index X l (t) drops to the device maintenance point M p , the device is maintained. The service quality after device maintenance usually cannot be fully restored to the initial state. Therefore, the state index Z l < QoS0, where QoS0 = Q0 or R0, representing the initial quantity. The improvement factor is used to describe the improvement of the state index after device maintenance. The calculation formula for the improvement factor is (X l (t) - Z l ) / X l (t), where X l (t) is the device state index before maintenance. As the number of maintenance times increases, the service quality after maintenance will continue to decrease. When the state index after maintenance drops to QT u , it needs to be updated, and the state index after update is restored to QoS0.

[0112] Step 1.2: Establish a cloud manufacturing logistics service network model. The components of the model are shown in Table 2:

[0113] Table 2 Components of the cloud manufacturing logistics service network model

[0114]

[0115] Among them, the logistics service cost is: LS C = Cc * LS S * Dis, where Cc is the conversion coefficient between the transfer speed and the transfer cost, and Dis is the distance;

[0116] Step 1.3: Establish a directed graph model for cloud manufacturing tasks:

[0117] After the complex manufacturing task is divided, the association relationship between its various manufacturing subtasks can be represented by a directed graph model. The directionality of the edges in the directed graph represents the constraint relationship between manufacturing tasks, that is, from the predecessor task to the successor task. In addition to the execution constraint relationship between cloud manufacturing subtasks, each cloud manufacturing task has a specified delivery period. The functional requirements and basic attributes such as the workload of cloud manufacturing subtasks are shown in Table 3.

[0118] Table 3 Functional requirements and basic attributes of cloud manufacturing subtasks

[0119]

[0120] Step 2: Establish an integrated optimization model for cloud manufacturing service composition and scheduling;

[0121] The cloud manufacturing service platform completes customers' complex manufacturing needs by configuring manufacturing resources, while also meeting customers' constraints in terms of quality, cost, delivery time, etc. Three optimization goals of the cloud manufacturing service combination and scheduling integration optimization problem are set from the three perspectives of QoS, service cost, and on-time delivery.

[0122] 1.QoS includes two dimensions: reliability and quality, as shown in the following formula:

[0123] QoS = k1PS Qu +k2PS Re

[0124] k1+k2=1

[0125] TCQoS=∑QoS i

[0126] PS Qu ,PS Re are the quality and reliability of the cloud manufacturing service solution, respectively. k1 and k2 are weight coefficients, which can be 0.5 respectively. TCQoS is the sum of the QoS of all cloud manufacturing tasks.

[0127] 2. Service cost C = PST c +LST c , where PST c The cost of production services, LST c For the logistics service cost, the optimization is to take the cost TC of all cloud manufacturing tasks as the optimization target.

[0128] 3. The time difference between the delivery time and the completion time of the cloud manufacturing task is used as the consideration indicator for on-time delivery. The specific working hours are calculated as Where I is the number of cloud manufacturing tasks, The delivery time of task i is generated by the cloud, is the actual completion time of cloud manufacturing task i. CT =T Pro +T Log +T Idl , where T Pro is the cloud manufacturing production task time, T Log Time for cloud manufacturing logistics service, T Idl The waiting time between task executions.

[0129] In summary, the cloud manufacturing service composition and scheduling integrated optimization includes three optimization objectives, namely, maximizing QoS, minimizing cost, and minimizing delivery time difference, as shown in the following formula:

[0130]

[0131] The constraints of the proposed cloud manufacturing task combination and scheduling integrated optimization model are as follows:

[0132] They respectively indicate that only one production / logistics subtask can be assigned to one production / logistics service at a time, i is the cloud manufacturing task index, j is the cloud manufacturing subtask index, m is the cloud manufacturing production service index, and N p is the total number of cloud manufacturing production services, k is the cloud manufacturing logistics service index, N l Total number of cloud manufacturing logistics services

[0133] PZ ijm =1 indicates cloud manufacturing production subtask ST ij is assigned to the mth production service, otherwise it is 0;

[0134] LZ ijk =1 indicates cloud manufacturing logistics subtask LZ ij is assigned to the kth logistics service, otherwise it is 0;

[0135] Indicates that only one production service can be assigned to process a cloud manufacturing task at the same time, N t represents the number of cloud manufacturing tasks, represents the number of subtasks within the i-th cloud manufacturing task;

[0136] max{PE i′j′m ,LE ijk}≤PS ijm , j′,m=k means that the task start time needs to be after the completion of other tasks and the preceding logistics tasks; PS ijm Perform subtask ST for the mth production service ij Start time, PE i′j′m Perform subtask ST for the mth production service i′j′ End time, LE ijk For the k-th logistics service, perform the subtask LE ij End time of

[0137] PE ijk ≤LS ijm , k = m-1, indicating that the logistics task needs to be started after the immediate production task is completed;

[0138] Constraint LZ ijk With PZ ijk is a binary decision variable.

[0139] Step 3: Use a multi-population co-evolutionary algorithm based on adaptive meta-knowledge transfer to solve the optimization model;

[0140] As a preferred solution, the iterative framework of multi-population co-evolution algorithm based on adaptive meta-knowledge transfer is as follows Figure 3 The step 3 comprises the following steps:

[0141] Step 3.1: Individual coding. The genetic coding content of a single individual includes the production task allocation code, the logistics task allocation code and the production task relative priority code, all of which are represented by real numbers between [0,1]. The real number codes of the production and logistics task allocation part are used to select services from the candidate service set for executing cloud manufacturing subtasks. The production task relative priority code is used to determine the scheduling priority of cloud manufacturing production tasks, and is also represented by real numbers between [0,1]. Specifically, the corresponding decoding steps include the following three steps:

[0142] 1. Production service selection and scheduling sequence determination. The number of candidate manufacturing service sets is determined according to the functional requirements of the manufacturing service. The number of candidate service sets multiplied by the corresponding code is the actual number of selected cloud manufacturing services. At the same time, the scheduling sequence is generated by reordering the relative priority of the production services.

[0143] 2. Logistics service selection. Determine the number of candidate logistics services based on the coordinate information of adjacent production services, also obtained based on the coding value;

[0144] 3. Scheduling plan generation: According to the selected cloud manufacturing service capability information, the cloud manufacturing task is inserted into the service provider's local production scheduling plan, a scheduling plan is generated, and the corresponding service quality, service cost, and delivery time difference are calculated.

[0145] It is not difficult to find that based on the optimization objectives of the established optimization model, the final service quality, service cost and delivery time difference of the scheduling plan reflect whether the individual has excellent fitness.

[0146] Step 3.2: Population initialization and subpopulation division

[0147] This example uses the Scramble randomized Halton algorithm and random generation method to initialize the population. Secondly, the K-means mean clustering algorithm is used to divide the initialized population into S n Each subpopulation iteratively searches the solution space from different regions, fully exploring the potential excellent solutions in the feasible domain of the problem and avoiding the population from falling into the local optimum.

[0148] Step 3.3: Subpopulation iterative evolution

[0149] In this example, each subpopulation adopts NSGA-Ⅲ or MOEA-DE for iterative evolution, the probability of selecting the two algorithms is 50%, and the number of iterative evolutions is 15. Then the meta-knowledge adaptive migration is performed, and then the next round of 15 iterative evolutions is performed. In other embodiments, other well-known iterative evolutionary algorithms can also be applied, and the selection probability and the number of iterative evolutions between algorithms can also be further adjusted.

[0150] Step 3.4: Meta-knowledge adaptive transfer and population update

[0151] The process of meta-knowledge extraction of sub-populations, adaptive meta-knowledge migration between sub-populations and population updating is as follows: Figure 4 Calculate the source population SP So With the target population SP Ta Distribution Center, In the formula is the value of the center of subpopulation k in dimension j; is the encoding value of the i-th individual of subpopulation k in dimension j; NP is the number of subpopulations; secondly, calculate the distance from the individual to the center of the subpopulation to which it belongs in each dimension: is the distance from the encoding value of the i-th individual in the k-th subpopulation in the j-dimension to the subpopulation center, is the encoding value of the i-th individual in the k-th subpopulation on the j-dimension. The subpopulation meta-knowledge contains the method and pattern information of searching for high-quality solutions in the subpopulation evolution process. The propagation of this information among subpopulations helps subpopulations to find potential high-quality solutions more efficiently.

[0152] Step 3.4.1 Meta-knowledge Adaptive Transfer

[0153] Step 3.4.1.1 traverses each sub-population and calculates the probability distribution model of the sub-population meta-knowledge; this example uses multivariate normal distribution to fit the probability distribution model of the individual coding values ​​in the sub-population, but is not limited to this, and there are:

[0154]

[0155] Where N is the number of individuals in the subpopulation, x i,j is the value of the i-th individual in the subpopulation on the j-th dimension, u j is the mean of the encoding values ​​of all individuals in the subpopulation in dimension j, is the standard deviation of the coded values ​​of all individuals in the subpopulation in the dimension; ∑ is the covariance matrix of the subpopulation; μ=(μ1,μ2,…,μ l )

[0156] Step 3.4.3.2 traverses each sub-population, and tries to calculate the cross entropy between the meta-knowledge probability models between each sub-population to obtain the sub-population cross entropy matrix; the cross entropy calculation formula between two sub-populations is as follows:

[0157]

[0158] P(x i ): Multivariate Gaussian distribution function of population 1

[0159] Q(x i ): Multivariate Gaussian distribution function of population 2

[0160] Step 3.4.3.3 Calculate the meta-knowledge adaptive migration threshold point based on the mean of all cross entropies in the cross entropy matrix, compare the cross entropy between sub-populations with the threshold point, and perform meta-knowledge adaptive migration. Specifically, the meta-knowledge adaptive migration threshold point is λ = 0.4 times the mean of the sub-population cross entropy matrix, but is not limited to this. If the cross entropy between sub-populations is greater than this value, the sub-populations perform meta-knowledge migration to generate new individuals according to the following formula: Where c Ta is the meta-knowledge of the source population, i.e., an i*j real number matrix, i is the number of individuals, j is the number of individual encoding dimensions. mk is the population center in the target population, i.e., a j-dimensional real number matrix.

[0161] Step 3.5 evaluates the new individuals generated by the adaptive migration of meta-knowledge and updates the individuals in the sub-population to keep the population size unchanged;

[0162] The new individuals obtained by adaptive transfer of meta-knowledge are evaluated. In this example, the best 2% of them are exchanged with the 2% with the worst fitness values ​​in the original population. In some other embodiments, the ratio is not limited to 2%.

[0163] Step 3.6: Iterate condition judgment,

[0164] In this example, the iteration is stopped when the number of iterations reaches 1000. In some other embodiments, other iteration termination conditions may be set, such as the fitness function reaching a certain magnitude, or the increment rain threshold of the maximum fitness function.

[0165] Test Example 1:

[0166] To verify the effectiveness of the proposed algorithm, the AMTMCA algorithm in this example is compared with the NSGA-Ⅲ, MOEA-RVEA and MOEA-DE algorithms. They are tested on the test set DTLZ1-7 respectively, and their performance on the indicators GD and HV is compared. The test results are shown in Table 4.

[0167] Table 4 Comparison of test indicators of different algorithms on the test set DTLZ1-7

[0168]

[0169]

[0170] From the GD values ​​in the table, we can see that the proposed algorithm has good convergence on most test functions, which is better than the other three algorithms, and its convergence performance on different test sets is stable. From the HV values ​​in the table, we can see that the proposed algorithm performs better in population uniformity and diversity, and is very stable.

[0171] Test Example 2:

[0172] The case set is generated using random numbers, and the ranges of the key parameters are shown in Table 5:

[0173] Table 5 Key parameter ranges

[0174]

[0175] It is worth noting that in this example, quality and reliability are set to share the same QoS parameters.

[0176] Randomly generated relevant data:

[0177] The cloud manufacturing task table is shown in Table 6, and the cloud manufacturing task directed graph is shown in Figure 6 shown.

[0178] Table 6 Cloud manufacturing task list

[0179]

[0180] The cloud manufacturing production resource table is shown in Table 7, and the matching relationship network between cloud manufacturing production resources and cloud manufacturing production subtasks is shown in Figure 6 shown.

[0181] Table 7

[0182]

[0183]

[0184] An integrated optimization model for cloud manufacturing service combination and scheduling is established. The proposed multi-population co-evolutionary algorithm and three algorithms, NSGA-Ⅲ, MOEA-RVEA and MOEA-DE, are used for comparison. The solution results are shown in the figure. Figure 7As shown in the figure, the solution set HV value of 3.46 obtained by the proposed algorithm is significantly greater than 1.43 of NSGA-Ⅲ algorithm, 0.51 of RVEA and 0.01 of MOEA-D. This shows that the proposed algorithm has good uniformity and diversity. On the other hand, after merging all the above populations and recalculating the Pareto frontier, the proportion of individuals in various populations in the global population in the Pareto frontier level is shown in the figure below. Figure 8 As shown in Figure 2, all solutions obtained by the proposed algorithm are the Pareto frontiers of the global population.

[0185] The above is only an embodiment of the present invention, and the common knowledge such as the known specific technical solutions and / or characteristics in the solution is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several modifications and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by the present invention shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A cloud manufacturing service combination and scheduling integrated optimization method, characterized in that: The steps include: Step 1: Establish a complex matching network of cloud manufacturing services and cloud manufacturing tasks, where cloud manufacturing services are divided into cloud manufacturing production services and cloud manufacturing logistics services, and cloud manufacturing tasks also include cloud manufacturing production tasks and cloud manufacturing logistics tasks; Step 2: Based on the complex matching network, with the optimization objectives of maximizing overall QoS, minimizing cost and minimizing delivery time difference, a cloud manufacturing service combination and scheduling integrated optimization model is established, wherein the QoS of each cloud manufacturing service in the network is quantified by the weighted sum of service quality parameters and service reliability parameters, both parameters are described dynamically in the time domain, and the initial value is the evaluation value obtained in advance; Step 3: Use a multi-population co-evolutionary algorithm based on meta-knowledge adaptive migration to solve the integrated optimization model, obtain a multi-objective optimization Pareto solution set, and decode it into an optimal scheduling solution.

2. The method according to claim 1, characterized in that The complex matching network includes multiple production resources, each of which corresponds to a production service. The parameters describing the production resources include: production resource number PR id , Provider PR P 、Function type PR FU , position coordinates [PR x ,PR y ]、Production Capacity PR t ; Parameters describing the production service include, production service number PS id , Reliability PS Re , Quality PS Qu , Cost PS C ; It also includes multiple logistics service providers and logistics services. The parameters describing the logistics service providers include the label LR id , Logistics service range radius LR RA , logistics service node set LR TP and logistics transfer speed LS S ; The parameters describing logistics services include: id , Cost LS C and service mode LS SP .

3. The method according to claim 2, characterized in that Reliability PS Re and quality PS Qu The time domain dynamic description is as follows: Let the initial service quality of the new equipment be Q0, and the service quality degradation curve is described by the Gamma process. The shape parameter α and size parameter β are fitted by the manufacturing resource curve data, and the Gamma process is used to describe the degradation behavior of the manufacturing resource service quality, as shown in the following formula: Where Γ(·) is the Gamma function; The real-time dynamic cloud manufacturing resource service quality is expressed by the device status index for service quality, that is, the initial service quality minus the degradation amount, as shown below: X l (t)=Q0-∫0 t f(t,β,α)dt; Similarly, with the initial reliability of the new equipment as R0 and the reliability degradation curve described by the Gamma process, we can get the equipment status index for reliability, which is used to represent the real-time dynamic reliability of the equipment. When the manufacturing service reliability is reduced to a certain level, it also needs to be maintained and replaced. When the device status index X l (t) drops to the device maintenance point M p , perform device maintenance on it. The status index after device maintenance is Z l <QoS0, where QoS0 = Q0 or R0, representing the initial quantity; use an improvement factor to describe the improvement of the status index after device maintenance. The calculation formula for the improvement factor is (X l (t) - Z l ) / X l (t), X l (t) is the device status index before maintenance. When the status index after maintenance drops to QT u , then it needs to be updated, and the status index after update is restored to QoS0.

4. The method according to claim 3, characterized in that The optimization objective is expressed as: Among them, TCQoS is the overall network QoS, which is calculated as follows: QoS=k1PS Qu +k2PS Re k1+k2=1 TCQoS=∑QoS i PS Qu ,PS Re The value range of is [0.5,1], k1, k2 are weight coefficients; TC is the cost of all cloud manufacturing tasks, and the cost of a single service C = PST c +LST c , where PST c The cost of production services, LST c is the logistics service cost, including: PS C =C Fle *C Mul *RS Re *PS Qu *PS Qu *C fun *Workload Among them, C fun is the unit time cost corresponding to each service type, C Fle The coefficient corresponding to the service flexibility type, C Mul Indicates the coefficient corresponding to the service function type, and Workload is the workload; as well as, LS C =Cc*LS S *Dis, Where CC is the conversion factor between transit speed and transit cost, and Dis is the distance; DTD is the delivery time difference, which is: Where I is the number of cloud manufacturing tasks, The delivery time of task i for cloud manufacturing, The actual completion time of cloud manufacturing task i is: T CT =T Pro +T Log +T Idl , Where T Pro is the cloud manufacturing production task time, T Log Time for cloud manufacturing logistics service, T Idl The waiting time between task executions.

5. The method according to claim 4, characterized in that The constraints of the optimization model include: They respectively indicate that only one production / logistics subtask can be assigned to one production / logistics service at a time, i is the cloud manufacturing task index, j is the cloud manufacturing subtask index, m is the cloud manufacturing production service index, and N p is the total number of cloud manufacturing production services, k is the cloud manufacturing logistics service index, N l The total number of cloud manufacturing logistics services, PZ ijm =1 indicates cloud manufacturing production subtask ST ij is assigned to the mth production service, otherwise it is 0, LZ ijk =1 indicates cloud manufacturing logistics subtask LZ ij is assigned to the kth logistics service, otherwise it is 0; Indicates that only one production service can be assigned to process a cloud manufacturing task at the same time, N t represents the number of cloud manufacturing tasks, represents the number of subtasks within the i-th cloud manufacturing task; Indicates that the task start time needs to be after the completion of other tasks and the preceding logistics tasks; PS ijm Perform subtask ST for the mth production service ij Start time, PE i′j′m Perform subtask ST for the mth production service i′j′ End time, LE ijk For the kth logistics service, perform the subtask LE ij End time of This indicates that the logistics task needs to start executing after the immediate preceding production task is completed; Constraint LZ ijk With PZ ijk is a binary decision variable.

6. The method according to claim 1, characterized in that Step 3 includes: Step 3.1: Individual coding; The genetic coding content of a single entity includes the production task allocation code, the logistics task allocation code and the production task relative priority code, all of which are represented by real numbers between [0,1]. The real number codes of the production and logistics task allocation parts are used to select services from the candidate service set to perform cloud manufacturing subtasks; The relative priority code of the production task is used to determine the scheduling priority of the cloud manufacturing production task, and is also represented by a real number between [0,1]; The corresponding decoding steps include: Determine the number of candidate manufacturing service sets according to the manufacturing service function requirements, multiply the number of candidate service sets by the corresponding code to obtain a selected cloud manufacturing service, and reorder the production services according to their relative priorities to generate a scheduling sequence; Determine the number of candidate logistics services based on the coordinate information of adjacent production services, also obtained based on the coding values; According to the selected cloud manufacturing service capability information, cloud manufacturing tasks are inserted into the service provider's local production scheduling plan, a scheduling plan is generated, and the corresponding service quality, service cost, and delivery time difference are calculated; Step 3.2: Initialize the population and divide it into S n subpopulation; Step 3.3: Each sub-population selects different algorithms according to probability and iterates them a certain number of times; Step 3.4: Meta-knowledge adaptive transfer and population update Calculate the source population SP So With the target population SP Ta Distribution Center: In the formula is the value of the center of subpopulation k in dimension j; is the encoding value of the i-th individual of subpopulation k in dimension j; NP is the number of subpopulations; Calculate the distance of each dimension from the individual to the center of the subpopulation to which it belongs: is the distance from the encoding value of the i-th individual in the k-th subpopulation in the j-dimension to the subpopulation center, is the encoding value of the i-th individual in the k-th subpopulation on the j-dimension; Step 3.4.1 Meta-knowledge Adaptive Transfer Step 3.4.1.1 traverse each sub-population and calculate the sub-population meta-knowledge probability distribution model; Step 3.4.3.2 traverses each subpopulation, calculates the cross entropy between the meta-knowledge probability models of each subpopulation, and obtains the subpopulation cross entropy matrix; the cross entropy calculation formula between two subpopulations is as follows: P(x i ): multivariate Gaussian distribution function of population 1, Q(x i ): multivariate Gaussian distribution function of population 2; Step 3.4.3.3 Calculate the meta-knowledge adaptive migration threshold point based on the mean of all cross entropies in the cross entropy matrix. If the cross entropy between sub-populations is greater than the threshold point, the sub-populations will perform meta-knowledge migration between each other. The migration process generates a new individual InD according to the following formula: new : Where c Ta is the meta-knowledge of the source population, i.e., an i*j real number matrix, i is the number of individuals, j is the number of individual encoding dimensions, and mk is the population center in the target population, i.e., a j-dimensional real number matrix; Step 3.5 evaluates the new individuals generated by the adaptive migration of meta-knowledge and updates the individuals in the sub-population to keep the population size unchanged; The new individuals obtained by adaptive transfer of meta-knowledge are evaluated, and the best proportion of them are exchanged with the proportion of the worst fitness value in the original population. Step 3.6: Determine the iteration stop condition. If it is met, stop the iteration. If not, return to step 3.3 for the next round of iterative evolution.

7. The method according to claim 6, characterized in that In step 3.2, the Scramble randomized Halton algorithm and random generation method are used to initialize the population.

8. The method according to claim 6, characterized in that In step 3.2, the K-means clustering algorithm is used to divide the initialized population into sub-populations.

9. The method according to claim 6, characterized in that In step 3.3, each subpopulation uses NSGA-Ⅲ or MOEA-DE for iterative evolution, and the probability of selecting the two algorithms is 50%.

10. The method according to claim 6, characterized in that The threshold point of meta-knowledge adaptive migration is taken as λ < 1 times the mean of the sub-population cross entropy matrix.

Images