A service recommendation method for production factors in industrial Internet environment

Abstract

The present invention provides a service recommendation method for production factors in an industrial Internet environment. First, for each decomposed production task, two algorithms are used to screen the historical data of production factor service collaboration to obtain frequent items and similarity information, that is, a similarity algorithm based on SimRank and a frequent item mining algorithm based on FP‑growth (Frequent Pattern Growth) are used to find out the hidden collaboration effectiveness information between services; then, by quantifying the collaboration effectiveness index and combining the MOEA (Multi‑objective Evolutionary Algorithm) algorithm, high-quality and high-collaboration effectiveness composite services are screened for users, thereby improving the service quality of the industrial Internet platform. Taking actual production tasks as an example, the present invention has demonstrated the effectiveness of the method through experiments conducted on simulation data sets, which can effectively improve the performance of service recommendations.

Description

Technical Field

[0001] The present invention mainly relates to an industrial Internet service recommendation method, specifically a service recommendation method for industrial Internet composite services that refers to and quantifies historical collaborative relationships. Background Art

[0002] With the continuous development of new IT technologies such as distributed computing, artificial intelligence, the Internet of Things, and big data, manufacturing resources can be perceived from the physical world and shared in the information world. In the context of the deep integration of new generation information technology and manufacturing technology, driven by the needs of industrial digitalization, networking, and intelligent transformation, the industrial Internet, which is characterized by comprehensive interconnection, comprehensive perception, intelligent optimization, and security and stability of various production factors such as people, machines, and objects, has come into being. In the industrial Internet environment, manufacturing enterprises virtualize manufacturing resources into various manufacturing services and deliver them to the industrial Internet platform. With the emergence of the service-oriented manufacturing model, the industrial Internet platform can configure the required production factor services to complete complex production processes. These production factor services are collectively referred to as manufacturing services in the present invention. The industrial Internet platform can effectively help manufacturing users or ordinary users obtain the required manufacturing services and assemble them to perform the complex manufacturing tasks that need to be completed. As more and more manufacturing companies join the industrial Internet platform, more and more manufacturing services are published to the platform, making the selection of manufacturing services difficult. Therefore, using appropriate methods to recommend suitable manufacturing service combinations for users' production process needs plays a vital role in improving the service recommendation efficiency and service quality of the industrial Internet platform, and is also an effective way to solve the service selection problem.

[0003] Data mining is a new type of technology based on big data, and its purpose is to mine valuable information from the amount of data that cannot be processed manually in a short period of time. Composite services that collaborate well on the Industrial Internet platform will have a better chance of being successfully called and becoming frequent items in historical data. Therefore, data mining for frequent items can discover and recommend highly collaborative service combinations. When evaluating the performance of service combinations, due to the existence of multi-dimensional QoS indicators, some manufacturing tasks have many parallel solutions, which may not appear in historical data, so the collaborative effectiveness of certain service combinations cannot be evaluated only by data mining. To alleviate this problem, the present invention attempts to combine graph-based similarity calculation and frequent item mining to obtain the similarity relationship between composite services, and then evaluate the effectiveness of service collaboration.

[0004] So far, a lot of work has been done on service recommendation. However, the existing research work mainly follows the service composition ideas in the traditional service computing field, selects a manufacturing service for each task in the composition process to complete its task, or realizes service selection through the combination of optimization models and heuristic algorithms. The actual algorithm is driven by the actual problem. In different manufacturing scenarios, it is necessary to redesign the relevant algorithms according to the particularity of the optimization problem so that the algorithm can obtain a more satisfactory solution from the application perspective. Summary of the invention

[0005] In order to solve the shortcomings of current technology, the present invention combines the existing technology and starts from actual production examples to try to solve the problems in the recommendation of industrial production factor services for different types of services, so as to effectively improve the production quality and efficiency of the industrial Internet platform.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] 1. A service recommendation method for production factors in the industrial Internet environment. The process includes the following steps: (1) Task decomposition: decompose the user's production tasks; (2) Data processing: use SimRank and FP-growth to filter frequent items in historical data and obtain similarity data between services; (3) Quantify indicators and combine with MOEA: combine the MOEA algorithm with the collaboration effectiveness indicator CEI (Collaboration Effectiveness Indicator) to filter high-quality and high-collaboration effectiveness composite services; (4) Recommendation: The selected composite services are combined into a composite service and delivered to the user to complete the production task.

[0008] 2. According to the service recommendation method for production factors in an industrial Internet environment as described in claim 1, in the step (1), the tasks are pre-processed when decomposing the production tasks, including the QoS indicators of the tasks, the resources required to complete the services, including production resources, the economic costs of material resources, etc., and the task collaboration data of the industrial Internet platform.

[0009] 3. According to the service recommendation method for production factors in an industrial Internet environment as described in claim 1, in the steps (2) and (3), the steps of the algorithm are:

[0010] (1) Use Python to read the call relationship of the composite service from the selected historical collaboration data;

[0011] (2) Use the SimRank algorithm to generate a relationship matrix from the call relationship;

[0012] (3) Iterate the relationship matrix to calculate the similarity between all services;

[0013] (4) Use the FP-growth algorithm to filter frequent items in the composite service collaboration history;

[0014] (5) Combine frequent items and SimRank similarity to evaluate the collaborative frequency index CEI of composite services;

[0015] (6) In each iteration of MOEA, individuals with high CEI are prioritized to simultaneously screen out composite services with high CEI and high QoS.

[0016] (7) Multiple optimized composite services form a recommendation list and recommend it to users.

[0017] 4. According to a service recommendation method for production factors in an industrial Internet environment as described in claim 3, it is characterized in that in the step (1), the industrial Internet platform should have rich collaborative relationship data, including but not limited to industrial production records, service collaboration records and other structured and unstructured information.

[0018] 5. A service recommendation method for production factors in an industrial Internet environment according to claim 1, characterized in that, in the step (3), before the algorithm is processed, the data has been formatted and adapted to the input of the algorithm;

[0019] 6. According to the service recommendation method for production factors in an industrial Internet environment as described in claim 3, it is characterized in that in the step (3), the parameter adjustment of the MOEA algorithm uses the default parameters recommended by the algorithm, and the results are standardized.

[0020] Beneficial effects of the present invention:

[0021] In the present invention, by combining the service collaboration history information, the SimRank similarity and the frequent items of FP-growth are combined to evaluate the collaboration effectiveness of the service. At the same time, the collaboration effectiveness index is combined with the multi-objective evolutionary algorithm to screen candidate services. Finally, the experiment shows that the HV, mean f, CEI and other indicators of the composite service obtained by the present invention are significantly improved compared with the traditional MOEA classification method. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a sample diagram of the ship production process of the present invention;

[0023] Figure 2 A result of filtering the composite service to form a composite service;

[0024] Figure 3 Bipartite graph serving the task;

[0025] Figure 4 Service recommendation process overall model framework diagram

[0026] Figure 5 It is the flow chart of MOEA+CEIF algorithm;

[0027] Figure 6 The results of the filter ratio P in the orthogonal experiment of MOEA+CEIF algorithm parameter selection;

[0028] Figure 7 The result of frequent item threshold F in the orthogonal experiment of MOEA+CEIF algorithm parameter selection;

[0029] Figure 8 This is a comparison chart of the CEI index in MOEA+CEIF and the two original MOEA methods;

[0030] Fig. 9 This is a comparison chart of the HV index in MOEA+CEIF and the two original MOEA methods;

[0031] Fig.10 This is a comparison chart of the mean f indicator in MOEA+CEIF and the two original MOEA methods. DETAILED DESCRIPTION

[0032] In conjunction with the accompanying drawings and specific embodiments, the present invention will be further described. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art may make various changes or modifications to the present invention, and these equivalent forms are also within the scope limited by the present application.

[0033] Since the existing research work mainly follows the service combination idea in the traditional service computing field, selects a manufacturing service for each task in the combination process to complete its task, or realizes service selection through the combination optimization model and heuristic algorithm. To this end, the present invention proposes a new recommendation method for production factor services for industrial Internet platforms, called MOEA+CEIF. We first process the historical collaboration data of the industrial Internet platform based on SimRank and calculate the similarity between services. Subsequently, the FP-growth frequent item mining algorithm is used to screen the composite services and quantify the collaboration effectiveness index. Secondly, MOEA is combined with the collaboration effectiveness index, namely MOEA+CEIF, to recommend composite services that meet the needs of users. Finally, the composite service with both high collaboration effectiveness and QoS indicators is delivered to users.

[0034] The overall framework of the production factor service combination recommendation method based on collaboration-driven is as follows: Figure 1As shown in the figure, it includes four main parts: 1) Task decomposition: Decompose several production tasks based on user needs to form a task chain; 2) Historical data processing: Based on SimRank, describe the collaboration similarity between services in the historical records, and then use the FP-growth algorithm to find frequent items to mine and quantify the collaboration effectiveness hidden in the collaboration history; 3) Production factor service combination optimization algorithm: Combine the collaboration effectiveness index and the MOEA algorithm to find composite services with both high QoS index values ​​and high collaboration effectiveness index values; 4) Service recommendation: Multiple optimized composite services are formed into a composite service and recommended to users.

[0035] The examples, model construction and experiments of the present invention are described in detail below.

[0036] Industrial production example description:

[0037] In order to describe the problem of production factor service combination in the industrial manufacturing link under the industrial Internet environment, this paper takes a coarse-grained ship production process in a real scenario as an example to illustrate its production process. Figure 2 As shown in the figure, the ship production process includes five different tasks: steel processing, section production, equipment processing, berth closure and ship testing. Each task requires at least one type of production factor resources, namely material resources or machine resources, and each type of production factor resources may require one or more. Task T1 requires two types of production factor resources and two types of each type of production factor resources are required. Task T3 requires two machine resources, and task T5 only requires one machine resource. The sequence relationship between tasks can be sequential, parallel, conditional or cyclic. There are only two execution relationships between subtasks in this ship production process: sequential and parallel. In this process, T1, T2, T3, and T4 all need to be completed by composite services, that is, the collaboration of two or more manufacturing services, while T5 only requires one manufacturing task to complete. Based on each task or composite service, a complete service combination chain can be formed. Assume that each manufacturing service has multiple candidate services, such as Figure 3 The following is a possible service combination solution.

[0038] 1. Recommendation of production factor service portfolio driven by collaboration

[0039] The comprehensive service quality of the service combination considered in the present invention includes time cost and economic cost, and its optimization objective function is shown in formula (1). c The economic cost Q of the service combination t is the time cost of the service combination, t0 in formula (2) is the maximum time cost constraint of a single service, and c0 is the maximum economic cost constraint of a single service.

[0040] f=min{Q c ,Qt} (1)

[0041]

[0042] The QoS indicators of services in the present invention are: (1) time, which refers to the sum of execution time required to complete the specified manufacturing service. (2) cost, which refers to the economic cost of executing the manufacturing task for the specified manufacturing service. There are four typical combination structures of QoS evaluation indicators: sequential, parallel, conditional and cyclic, as shown in Table 1.

[0043] Table 1 QOS optimization objectives

[0044]

[0045] 1.1 Frequent Item Mining Algorithm Based on FP-growth

[0046] A service combination that has been successfully called multiple times should theoretically have a higher CEI index value, and the service combination is more likely to be selected. Therefore, the frequent items mined from the historical data by FP-growth should theoretically have a higher CEI index value, that is, first select the service combination with a theoretically high CEI index. Then, the CEI index of the service combination of infrequent items is calculated in combination with the similarity between services. There are two important parameters when searching for frequent items: support (Sup) and confidence (Conf). In FP-Growth, it is necessary to set the minsup parameter of the algorithm, that is, the minimum support. Minsup and the frequent items of the service combination are determined by experiments and will be described in detail in the specific embodiments. The calculation method of Sup is shown in Formula 3, and the calculation method of Conf is shown in Formula 4 below.

[0047]

[0048]

[0049] Among them, N sc represents the number of times the combination is called, and N represents the total number of times the combination is called in history. x and Service Portfolio SC y The relationship is that the former needs to include the latter.

[0050] 1.2 Collaborative Similarity Calculation Based on SimRank

[0051] Each composite service in the frequent item set is a service combination that frequently appears in history and serves a certain task or a class of tasks. After distinguishing the frequent and infrequent service combinations through FP-Growth screening, for the same or different types of services, in order to find the hidden relationship between services, the SimRank algorithm can be used to calculate the similarity between services. This algorithm is a method based on graph theory to calculate the similarity between any two nodes. The basic idea of ​​the algorithm is: if two entities are similar, the individuals associated with them should be similar. Figure 4 As shown, the solid line represents the calling relationship between tasks and services. Among them, the hexagons represent different materials, which are the materials needed to complete a certain task, and the circles represent automated equipment such as production machines and automatic factories. The services called by the three tasks are: T1: S1, S2, S4, S6; T2: S1, S2, S4 and S7; T3: S2, S3, S5 and S7. Since the service combinations called by T1 and T2 are only different in S6 and S7, S6 and S7 in the combination are theoretically very similar. From the service combinations called by T2 and T3, it can be found that there is also a correlation between S1, S2, S4 and S5, but it should be less than the correlation between S6 and S7.

[0052] Based on the historical call relationship between services and tasks described in the dataset, we can iterate to obtain the similarity and similarity matrix between tasks that have appeared in history. The iterative method is shown in Formula 5. According to relevant research, we can further determine the similarity between any two services through Formula 6.

[0053]

[0054]

[0055] In Formula 5, S represents the similarity matrix, Q represents the transition probability matrix, and the sum of each column of Q is 1, which means that if node i can be transferred to node j, and there are n such nodes i, then The coefficient a is the damping factor between 0 and 1. n is the n×n identity matrix. In Formula 6, SR(S x ,S y ) indicates S x and S y The similarity between x ) and I(S y ) indicates S x and S y The neighbor set of I i (S x ) and I j (S y ) indicates S x and Sy The damping coefficient a is the same as in Formula 6. |I(S x )| and |I(S y )| represents the set I(S x ) and I(S y ) is the number of elements in .

[0056] 1.3 Calculating Collaboration Effectiveness Indicators

[0057] After obtaining the similarity and frequent items, we can further calculate the writing effectiveness. Considering the characteristics and connections between similarity and frequent items, we intend to use similarity and frequent items to define and calculate the collaborative effectiveness index. The specific steps are as follows:

[0058]

[0059] In Formula 7, a is the damping coefficient in Formula 6, and F re The frequent item set filtered by FP-growth, N rj is the number of services in the jth frequent item, N i is the number of all combinations in the frequent item set, S xi For the combination S x For the service combination belonging to frequent items, S rj is the frequent item set F re If the composite service is a frequent item, the CEI index is directly related to the number of composite services. For composite services based on infrequent items, its index is directly related to the similarity of composite services belonging to frequent items.

[0060] 2. Production Factor Service Combination Optimization Algorithm

[0061] In this section, the present invention proposes a CEI-based filtering method MOEA+CEIF. The algorithm can be combined with MOEA to screen high CEI individuals in the population in iteration to improve the optimization effect. In order to ensure that the solution has a high comprehensive QoS and a high CEI index value, it is considered to combine the multi-objective optimization algorithm with the CEI evaluation method. Figure 5 This is the algorithm framework of MOEA+CEIF. In the process of screening and iteration of each generation, the focus is on retaining the business combination with high CEI, so that individuals with high CEI can be retained in the population evolution of each generation, so as to ensure that the Pareto optimal combination of QoS is obtained on the basis of the high CEI of the service combination during the calling process. The specific steps of the MOEA+CEIF algorithm are as follows:

[0062] (1) Set the number of population iterations, population size, crossover mutation probability, and filtering ratio parameters;

[0063] (2) Generate and read the population solution set P, and create an elite solution set P';

[0064] (3) Copy the population set P' to the set P' and calculate the number of dominated solutions for each element in P'. The number of dominated solutions for the i-th element is S(i), where S(i) is defined as shown in Formula 8.

[0065] (4) Delete the dominated solution set from P'. If the number of individuals in P' exceeds the maximum number of individuals N, prune the current P' through the clustering method. In the clustering method, cluster each element in P' to build a cluster set. When pruning, select clusters with smaller distances to merge, and select the cluster center as the reference point. The distance between clusters c1 and c2 is calculated as shown in Formula 9;

[0066] (5) Randomly select individuals from P∪P' for crossover and mutation to generate the next generation population, and then randomly select individuals for crossover and mutation;

[0067] (6) Calculate the CEI of each individual after crossover mutation, sort them using the CEI indicator under the same dominance, and filter the solution set according to a pre-set ratio;

[0068] (7) Select the population by the roulette method. In roulette, individual fitness is treated as a probability factor in each calculation. However, individual fitness F is affected by the distance metric D and the dominance relationship metric R. R(i) is the sum of S(j) of all individuals j that dominate i. D(i) is to ensure that when R(i) of two points is the same, there is a certain distance between the frontiers of two adjacent generations. The calculation is shown in formula 10-12;

[0069] (8) Determine whether the iteration has reached the maximum number of iterations. If it ends, return and set the service combination indicator triple recommended to the customer as {S t , S c , S ex}. Among them, S t is the optimal solution for QoS time cost, S c is the optimal solution of economic cost, S ex is the optimal solution for the mean;

[0070] S(i)={j|j∈PUP′,i>j} (8)

[0071]

[0072] R(i)=∑ j∈P∪P′ S(j) (10)

[0073]

[0074] F(i)=R(i)+D(i) (12)

[0075] In Formula 9, |c1| and |c2| represent the number of cluster centers, and |i1-i2| represents the Euclidean distance between two points. In Formula 11, D(i) is always between 0 and 1. It represents the distance from point i to the point k farthest from point i.

[0076] Example:

[0077] In this embodiment, an experimental verification is carried out for the recommendation method provided by the present invention. The data set and experimental parameter settings, evaluation indicators, comparison methods and experimental results of this embodiment are described in detail below.

[0078] Dataset and Experimental Settings:

[0079] We use a simulation data set for experiments. The data set is randomly generated using Python, including the historical call relationship between services and tasks in the random generation of service combinations and the execution time and cost of each service. 200 candidate services of two different types are generated, and their call relationships are randomly generated.

[0080] Parameter settings have a great influence on the experimental results. The present invention conducts multiple experiments on the generated simulation data set, attempting to determine the values ​​of some parameters of the algorithm in the simulation set through orthogonal experiments, as shown in Table 2. Each group of experiments is conducted 10 times and the average is taken as a reference. The results are shown in Table 2. Figures 6-7 As shown. Finally, the values ​​of all parameters are shown in Table III, where “-” means that the original MOEA does not need to set this parameter. The parameter selection algorithm in the comparison method recommends the default parameters, such as: PC is the crossover probability, PM is the mutation probability. Two parameters are required to calculate the effectiveness of the collaboration: the filtering ratio P and the frequent item threshold F, N is the population number, and IN is the number of iterations.

[0081] Table 2 Parameters of the experiment

[0082]

[0083] Table 3 Parameter setting table

[0084]

[0085] Evaluation Metrics:

[0086] QoS includes multiple optimization indicators, and it is generally difficult to measure the effect of the algorithm with some general evaluation criteria. In this experiment, the Hypervolume indicator evaluation method is proposed, which represents the volume of the hypercube enclosed by the individuals in the solution set and the reference point in the target space. Related studies have shown that if a solution set S is better than another solution set S', then the Hypervolume indicator of the solution set S will also be greater than the Hypervolume indicator of the solution set S', and Hypervolume can evaluate the convergence of the solution and the distribution of the solution together. The accuracy of calculating the Hypervolume indicator depends on the selection of the reference point, that is, when evaluating the same solution set, different reference points will be selected, and different calculation results will be obtained. If the true frontier is unknown, the reference point (1,1) can be selected. If the true Pareto frontier is known, the point closest to the origin can be selected, or the point with the best attributes can be selected as the reference point. This experiment intends to use (1,1) as the reference point, and the calculation method of the Hypervolume indicator is shown in Formula 15.

[0087]

[0088] Among them, X represents the non-optimal solution set of the algorithm, represents the set of all reference points, is the Hypervolume index of the non-optimal solution set X, Leb is the Lebesgue measure of X, [f1(x),r1]×[f2(x),r2]×…×[f N (x),r N ] indicates that the solution x is dominated but not Dominant individuals.

[0089] Comparison method:

[0090] In order to verify the algorithm performance of MOEA+CEI, a series of experiments were conducted. Compared with two classic MOEA algorithms: SPEA-II and NSGA-II, according to relevant research, SPEA-II and NSGA-II are better than traditional PSO and GA algorithms, so the present invention does not compare them. The present invention conducts an experimental comparison between MOEA+CEI and the original MOEA to verify the ability of the method of the present invention in screening composite services with high optimization performance and high CEI indicators, and demonstrates the overall performance and effect.

[0091] SimRank’s similarity (SR) is different from the method based on Pearson correlation coefficient (PCC) or cosine similarity that describes customer ratings. It is based on the call relationship between services rather than targeting a single service, so this algorithm is selected for similarity calculation. In Tables 4 and 5, the above two similarity calculations for the candidate composite services of task T1 show that s2 and s 103The SR between them is 0.70, which is higher than that between s2 and s 104 0.40, but s2 and s 103 The PCC is 0.58, which is smaller than s2 and s 104 The PCC value of s2 and s 103 The number of collaborations is greater than that of s2 and s 104 This phenomenon occurs because the frequency of collaboration can affect the similarity between services to a certain extent, and is not entirely determined by the objective attributes of the services. Therefore, the SimRank algorithm can more effectively and comprehensively evaluate the similarity between services.

[0092] Experimental results and analysis:

[0093] The similarity (SR) of SimRank is different from the method based on Pearson correlation coefficient (PCC) or cosine similarity that describes customer ratings. It is based on the call relationship between services rather than targeting a single service. Therefore, this algorithm is selected for similarity calculation. In Tables 4 and 5, the above two similarity calculations for the candidate composite services of instance task T1 show that s2 and s 103 The SR between them is 0.70, which is higher than that between s2 and s 104 0.40, but s2 and s 103 The PCC is 0.58, which is smaller than s2 and s 104 The PCC value of s2 and s 103 The number of collaborations is greater than that of s2 and s 104 This phenomenon occurs because the frequency of collaboration can affect the similarity between services to a certain extent, and is not entirely determined by the objective attributes of the services. Therefore, the SimRank algorithm can more effectively and comprehensively evaluate the similarity between services.

[0094] Table 4 SR similarity of candidate services for task T1 (partial)

[0095]

[0096] Table 5 PCC similarity of candidate services for task T1 (partial)

[0097]

[0098] Table 6 is the statistical results of the comparative experiment. In the experiment, each algorithm was run 10 times to avoid the randomness of the algorithm operation, and all results were standardized. The selected evaluation indicators are HV, mean f and CEI. Mean f represents the average value of all QoS items in the non-dominated solution set, which can numerically reflect the quality of the optimization results. It can be seen that the improvement of QoS indicators in the three indicators is basically the same. The improvement of CEI is more obvious in NSGA-II. After adding the CEI indicator, SPEA-II and NSGA-II have a significant improvement in collaborative effectiveness. The improvement of the HV indicator is basically the same in the two algorithms. In order to make the results more intuitive, the present invention uses box plots to display the experimental results, such as Figures 5 to 7 As shown, the points in the box plot are mean points, the line in the box plot is the median line, and the points outside the box plot are outlier points.

[0099] Table 6 Experimental results of the algorithm and the comparison algorithm

[0100]

[0101]

[0102] like Figures 8 to 10 As shown in the figure, adding the CEI indicator can improve the performance of the original MOEA. This is because there are hidden collaborative relationships between services due to reasons such as the relationship between enterprises. These hidden relationships cannot be discovered directly, but can be reflected through the call relationship between services. MOEA+CEI considers the collaborative relationship between services implicit in the composite service call history and utilizes valuable information in historical data. At the same time, composite services with high CEI tend to be retained during the selection process, making the results better than the original MOEA, which is more obvious in NSGA-II. It can be seen that the optimization performance improvement of each indicator of MOEA+CEI under SPEA-II is slightly smaller than that of NSGA-II. Compared with SPEA-II, the algorithm of the present invention is more suitable for NSGA-II. Therefore, when solving the composite optimization problem of manufacturing services, MOEA+CEI has better optimization performance than the original MOEA.

[0103] Aiming at the service combination demand for production factors under the industrial Internet, the present invention proposes a hybrid service recommendation method for collaboratively optimizing service collaboration effectiveness and service quality, namely MOEA+CEIF, by combining the classic MOEA and the service collaboration effectiveness index CEI. Specifically, the present invention models the service combination optimization problem as a dual-objective Pareto optimization problem. To solve this problem, the present invention combines the SimRank similarity algorithm and the FP-growth algorithm, applies the Pareto optimality idea of ​​the multi-objective optimization algorithm, solves the service combination optimization problem, and recommends service combination optimization solutions with high collaboration efficiency and optimized service quality to users. A series of experimental results show that compared with other benchmark methods, MOEA+CEI can ensure higher collaboration effectiveness while ensuring QoS indicators.

Claims

1. A service recommendation method for production factors in an industrial Internet environment, the process of which includes the following steps: (1) Task decomposition: Decompose the user's production tasks. Each production task requires at least one type of production factor resource, namely material resources or machine resources, and each type of production factor resource requires one or more types. Based on each task or composite service, a complete service combination chain is formed; (2) Data processing: Use SimRank and FP-growth to filter frequent items in historical data and obtain similarity data between services; (3) Quantitative indicators and combined with MOEA: Combine the MOEA algorithm with the collaboration effectiveness indicator CEI (Collaboration Effectiveness Indicator) to screen high-quality and high-collaboration effectiveness composite services; combine frequent items and SimRank similarity to evaluate the collaboration frequency indicator CEI of composite services; give priority to individuals with high CEI in each iteration of MOEA to simultaneously screen out composite services with high CEI and high QoS; Multiple optimized composite services form a recommendation list and recommend it to users; The QoS indicators of the service combination chain include: time cost and economic cost; The QoS indicator is evaluated by using a Hypervolume indicator evaluation method, wherein the calculation method of the Hypervolume indicator is: Among them, X represents the non-optimal solution set of the algorithm, represents the set of all reference points, is the Hypervolume index of the non-optimal solution set X, Leb is the Lebesgue measure of X, [f1(x),r1]×[f2(x),r2]×…×[fN(x),rN] represents the solution dominated by the solution x but not dominant individuals; In steps (2) and (3), the steps of the algorithm are: (1) Use Python to read the call relationship of the composite service from the selected historical collaboration data; (2) Use the SimRank algorithm to generate a relationship matrix from the call relationship; (3) Iterate the relationship matrix to calculate the similarity between all services; (4) Use the FP-growth algorithm to filter frequent items in the composite service collaboration history; Screening the high-quality and highly collaborative composite service includes: (1) setting the number of population iterations, the number of populations, the probability of crossover and mutation, and the filtering ratio parameter; (2) generating and reading the population solution set P, and creating an elite solution set P'; (3) copying the population set P' to the set P', and calculating the number of dominant solutions for each element in P', where the number of dominant solutions for the i-th element is S(i); (4) deleting the dominated solution set from P', and if the number of individuals in P' exceeds the maximum number of individuals N, pruning the current P' by a clustering method; in the clustering method, clustering is performed for each element in P' to construct a cluster set; when pruning, clusters with smaller distances are selected for merging, and the cluster center is selected as a reference point; (5) randomly selecting individuals from P∪P' for crossover and mutation After generating the next generation of population, randomly select individuals for crossover and mutation; (6) calculate the CEI of each individual after crossover and mutation, sort them with the CEI indicator under the same dominance, and filter the solution set according to a pre-set ratio; (7) select the population by the roulette method; in roulette, the individual fitness is regarded as a probability factor in each calculation; however, the individual fitness F is affected by the distance metric D and the dominance relationship metric R; and R(i) is the sum of S(j) of all individuals j that dominate i; D(i) is to ensure that when R(i) of two points is the same, there is a certain distance between the frontiers of two adjacent generations; (8) determine whether the iteration has reached the maximum number of iterations, and return the recommended service combination indicator triple to the customer at the end, and set it to {S t , S c , S ex }; where S t is the optimal solution for QoS time cost, S c is the optimal solution of economic cost, S ex is the optimal solution for the mean; (4) Recommendation: The screened composite services are formed into combined services and delivered to users to complete production tasks.

2. According to claim 1, a service recommendation method for production factors in an industrial Internet environment is characterized in that: In step (1), the industrial Internet platform should have rich collaborative relationship data, including but not limited to industrial production records, service collaboration records and other structured and unstructured information.

3. According to the service recommendation method for production factors in an industrial Internet environment of claim 1, it is characterized in that: In the step (3), data formatting has been completed before the algorithm is processed to adapt the input of the algorithm.

4. According to claim 1, a service recommendation method for production factors in an industrial Internet environment is characterized in that: In step (3), the parameters of the MOEA algorithm are adjusted using the default parameters recommended by the algorithm, and the results are standardized.

Images