Clustering methods

Abstract

The present invention provides a clustering method for grouping machines constituting a fluid processing device while being able to quantitatively evaluate the grouping results. The clustering method includes the following steps: creating a device model, wherein the device model represents the connection relationship between multiple machines (31) and pipes (4) constituting a processing device (1) in the form of a graph representing the connection relationship between nodes (51) and edges (52); setting clustering conditions for the device model; and using a computer to search for edges between clusters that meet the clustering conditions and have greater modularity, thereby clustering the device model.

Description

Technical Field

[0001] The present invention relates to a technology for grouping machines constituting a fluid processing facility. Background Art

[0002] The processing equipment for processing fluids includes natural gas equipment for liquefying natural gas or separating and recovering natural gas liquid, petroleum refining equipment for distilling or desulfurizing crude oil or various intermediate products, chemical equipment for producing petrochemical products or intermediate chemicals, polymers, etc. In addition, in this application, "fluid" refers to powders and particles (powders, particles, or granules) with fluidity in addition to gases and liquids.

[0003] These processing facilities have a structure in which a plurality of devices, such as static devices such as towers and heat exchangers and dynamic devices such as pumps and compressors, are arranged in a predetermined area and these devices are connected by pipes.

[0004] At this time, the arrangement of the plurality of devices constituting the processing facility is determined by comprehensively considering the topography and area constraints of the site, safety constraints, the usage amount (economic efficiency) of the structural materials of the devices and piping, and the like.

[0005] At this time, when arranging the machines constituting the processing equipment, if a plurality of machines are arranged in a disorderly manner, the pipes connecting the machines to each other become longer, and the total amount of pipe materials used increases. Therefore, sometimes the arrangement positions are determined in such a way that a plurality of machines are divided into a plurality of groups and the machines included in the same group are arranged in a concentrated area.

[0006] Previously, the following method was often used to group machines: for example, from the perspective of process design, a series of processes implemented in a processing device was divided into multiple groups, and the machines used in the divided processes were included in one group (for example, the module division corresponding to each processing block in Patent Document 1). On the other hand, when the content of fluid processing is abandoned and the connection relationship between the machine and the piping is focused on, it is not necessarily clear whether the grouping based on the previous method reflects the degree of relationship between the machines.

[0007] Therefore, a grouping method is required that can quantitatively grasp the grouping result of the determined devices while using the connection relationship between the devices and the pipes as a reference.

[0008] Prior art literature

[0009] Patent Literature

[0010] Patent Document 1: International Publication No. 2018 / 220703 Summary of the invention

[0011] Problems to be solved by the invention

[0012] The present technology provides a technology for grouping devices constituting fluid processing equipment while being able to quantitatively evaluate the grouping results.

[0013] Technical means of solving problems

[0014] This method includes the following steps:

[0015] Creating a facility model that represents a connection relationship between a plurality of machines constituting a processing facility for processing a fluid and pipes connecting the machines by means of a graph that represents a connection relationship between a node corresponding to each of the machines and an edge corresponding to each of the pipes;

[0016] For the device model, a clustering condition is set, wherein the clustering condition is at least one of the number of clusters to be clustered and the range of granularity, which is the sum of weights set on nodes included in each cluster; and

[0017] The device model is clustered by searching, by a computer, for edges between clusters that satisfy the clustering conditions and have greater modularity.

[0018] The clustering method may also have the following features.

[0019] (a) In the step of creating the equipment model, the equipment model is created such that the edges are weighted with a larger weight as the cost per unit length of the corresponding pipe becomes larger.

[0020] (b) In the process of making the equipment model, the equipment model is made as follows: for the node, the larger the occupied area of ​​the corresponding machine is, or the higher the pre-set evaluation value of the machine is, the greater the weight is.

[0021] (c) The clustering process is performed based on the Girvan-Newman algorithm.

[0022] (d) In (c), in the process of making the equipment model, the following equipment model is made: the greater the cost per unit length of the piping, the greater the number of unweighted edges connecting two nodes corresponding to two machines connected to the piping.

[0023] (e) In (c), the node corresponding to the machine included in the device model is an unweighted node; and in the process of making the device model, the following device model is made: the larger the occupied area of ​​the machine or the higher the pre-set evaluation value for the machine, the greater the number of connections of nodes, i.e., virtual nodes, which are connected to a node corresponding to the machine via an edge and not connected to any node other than the node.

[0024] (f) The clustering division process includes a process of counting the edges of the shortest path between two nodes included in the equipment model and adding a count value. When the processing equipment includes a piping branching at a branch point, in the process of making the equipment model, the following equipment model is made: a branch node that is not selected as the two nodes when the process of adding the count value is executed and is set corresponding to the branch point, and the branch node and the node corresponding to the machine connected to the branched piping are connected via an edge, thereby representing the branched piping.

[0025] Effects of the Invention

[0026] According to this method, a device model is created that represents the equipment and piping of the processing equipment as a graph including nodes and edges, and clustering of the device model is searched for in a manner that increases modularity under pre-set conditions. As a result, quantitative evaluation can be performed based on an evaluation criterion (modularity), and the grouping of the equipment constituting the processing equipment can be rationally implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a schematic diagram of the machine configuration of the processing equipment.

[0028] Figure 2 This is an example of a structure of a process including a device installed in a processing facility.

[0029] Figure 3 is a process model corresponding to the process.

[0030] Figure 4A This is the first explanatory diagram related to clustering.

[0031] Figure 4B This is a second explanatory diagram related to clustering.

[0032] Figure 4C This is the third explanatory diagram related to clustering.

[0033] Figure 4D This is the fourth explanatory diagram related to clustering.

[0034] Figure 5 This is an example of the configuration of a facility model that reflects the cost per unit length of piping.

[0035] Figure 6 This is an example of the structure of a facility model that reflects the occupied area of ​​the equipment, etc.

[0036] Figure 7 This is an example of the configuration of a device model that reflects the branching of piping.

[0037] Figure 8 is an example of a process for grouping based on clustering results.

[0038] Fig. 9 This is an example of the equipment configuration of a liquefied natural gas (LNG) facility.

[0039] Fig.10 is an example of a device model that is clustered.

[0040] Explanation of symbols

[0041] 1: Processing equipment

[0042] 10: Land

[0043] 2: Pipe rack

[0044] 3: Machine Group

[0045] 30: District

[0046] 31: Machine

[0047] 311a: Distillation Column

[0048] 311b: Distillation Column

[0049] 312a: Top Cooler

[0050] 313a: Receiver

[0051] 314: Liquid delivery pump

[0052] 4: Piping

[0053] 40: Branching point

[0054] 51: Node

[0055] 511: Dummy node

[0056] 512: Branch node

[0057] 52: Edge DETAILED DESCRIPTION

[0058] Figure 1 It is a plan view schematically showing the structure of a processing device 1 corresponding to a device model to which the clustering method of this example is applied.

[0059] As processing equipment 1 to which the present technology can be applied, the natural gas equipment, petroleum refining equipment, chemical equipment, etc. mentioned above can be exemplified. These processing equipment 1 include a plurality of machines 31 for processing fluids such as gases, liquids, and fluid powders and granules. Examples of the machines 31 provided in the processing equipment 1 include various processing towers such as distillation towers or reaction towers, static machines such as receiving tanks or heat exchangers for gas-liquid separation, and dynamic machines such as pumps or compressors. The machines 31 provided in the processing equipment 1 are interconnected via pipes 4 for the flow of fluids. In addition, in Figure 1 The piping 4 is omitted in the figure.

[0060] When constructing the processing facility 1 in the predetermined site 10, if a plurality of machines 31 are arranged in a disordered manner, the pipes 4 connecting the machines 31 to each other become longer, and the total usage or cost of the pipe materials increases. In addition, from the perspective of maintenance management or safety management, it is desirable to systematically arrange the machines 31 based on the predetermined idea.

[0061] Therefore, in a processing facility 1 provided with a plurality of machines 31, the machines 31 are grouped (hereinafter referred to as "machine groups 3") according to the order in which the fluid is processed or the relationship between the machines 31, and the machines 31 included in a common machine group 3 are sometimes arranged in a common partition area 30. Figure 1 A machine group 3 constructed in this way is recorded in .

[0062] In this example, the machine group 3 includes at least one machine 31 .

[0063] When the device group 3 includes a plurality of devices 31 , the pipes 4 connecting the devices 31 included in the common device group 3 are arranged in the common partition area 30 where the device groups 3 are arranged.

[0064] On the other hand, the pipes 4 connecting the machines 31 included in the different machine groups 3 are arranged, for example, so as to pass through the pipe rack 2. The pipe rack 2 is a frame structure that supports the pipes 4. In addition, the pipes 4 connecting the machines 31 included in the different machine groups 3 do not necessarily have to pass through the pipe rack 2. For example, between the machines 31 included in the machine group 3 in which the partition areas 30 are arranged adjacent to each other, the machines 31 may be connected by the pipes 4 that do not pass through the pipe rack 2.

[0065] like Figure 1As shown, by grouping a plurality of machines 31 provided in the processing facility 1 into a plurality of machine groups 3 and arranging the machines 31 included in each machine group 3 in a common partition area 30, for example, mutually related machines 31 can be arranged at positions close to each other. As a result, compared with a case where a plurality of machines 31 are arranged in a disorderly manner, it is possible to suppress the pipes 4 connecting the machines 31 from becoming excessively long, thereby suppressing the total amount of pipe material used.

[0066] On the other hand, as described above, in the processing facility 1 including the plurality of equipment groups 3 , it is not always clear which groupings are preferably performed when these equipment groups 3 are arranged in the site 10 .

[0067] Therefore, in this example, a device model corresponding to the processing device 1 is created, and a computer performs calculations to cluster the device model into a plurality of clusters. In this example, clustering of the device model refers to processing corresponding to grouping of the machines 31 included in the processing device 1.

[0068] When creating a facility model of the processing facility 1, data of a process flow diagram (PFD) created when designing the processing facility 1 can be used. The PFD describes the connection relationship between the devices 31 constituting the processing facility 1 through the pipes 4.

[0069] For example, Figure 2 It is a part of the process flow of the processing equipment 1 which simplifies the PFD. Figure 2 The following process is described: the gas flowing out from the top of the distillation tower 311a on the front stage is cooled and liquefied by an overhead cooler 312a, and a part of the liquid accumulated in the receiver 313a is sent to the distillation tower 311b on the rear stage by a liquid sending pump 314.

[0070] In the clustering method of this example, a facility model is created that expresses the connection relationship between the plurality of devices 31 constituting the processing facility 1 and the pipes 4 connecting the devices 31 , using a graph representing the connection relationship between nodes 51 and edges 52 .

[0071] like Figure 3 As shown, in the equipment model, the machines 31 constituting the processing equipment 1 correspond to the nodes 51 that can identify each other, and the pipes 4 connecting the machines 31 correspond to the edges 52 connected to these nodes 51. In addition, as described later, the branch point 40 of the pipe 4 can be represented by the node 51 (branch node 512).

[0072] When the PFD is computer-processable data, the connection relationship between the equipment 31 and the pipe 4 included in the PFD data can be read to create a device model. The device model is processed by a computer as data of an adjacency matrix or adjacency list representing the connection relationship between nodes 51 via edges 52.

[0073] The creation of the device model is not limited to a method that is automatically performed by a computer using a PFD. For example, data representing the device model may be manually input to a computer that performs clustering.

[0074] The device model expressing the connection relationship by the nodes 51 and the edges 52 can be divided into a plurality of clusters by a computer using a predetermined algorithm.

[0075] Examples of algorithms for clustering a network represented by a graph (device models in this example) include the Gervan-Newman algorithm that does not use an evaluation function (modularity, etc. described later) and the Newman algorithm that uses an evaluation function.

[0076] Figure 4A to Figure 4D Represents a procedure for clustering using the Girvan-Newman algorithm.

[0077] In the Girvan-Newman algorithm, the shortest path is determined for each pair of all nodes 51 contained in the network, and the edges 52 of the shortest path are counted. Figure 4A In the pair of nodes 51 indicated by hatching, the edge 52 indicated by the bold line constitutes the shortest path. Furthermore, a count value "+1" is added to each edge 52 (the process of adding a count value). Figure 4B There are two shortest paths in a pair of nodes 51 indicated by hatching in FIG. 1 . In this case, the value “+0.5” obtained by dividing the normal count value by the number of shortest paths becomes the count value of each edge 52 .

[0078] In this way, the same process is performed on all pairs of nodes 51 included in the network, and the count value of each edge 52 is added.

[0079] As a result, the count value of each edge 52 constituting the network becomes Figure 4C In this case, the new network obtained by cutting the edge 52 represented by the bold line with the largest count value is compared with the network obtained by using Figure 4A , Figure 4B The same processing as the example described is performed ( Figure 4D ).

[0080] In addition, for ease of explanation, Figure 4C , Figure 4D The numerical values ​​added to the respective figures are recorded virtually and are not the values ​​actually obtained by executing the Girvan-Newman algorithm.

[0081] exist Figure 4D In the network shown, when the edge 52 represented by the thick line with the largest count value is cut, two clusters, cluster 1 and cluster 2, are obtained. In other words, using Figure 4A to Figure 4D The described process performs a search for edges between clusters.

[0082] In this example, the network corresponds to the device model. After the device model is divided into a plurality of clusters, the devices 31 corresponding to the nodes 51 included in the common cluster are grouped so as to be included in the common device group 3.

[0083] Modularity is known as an index for evaluating the results of clustering by an algorithm. Modularity is defined by the following formula.

[0084] [Formula 1]

[0085] Q=∑ i (e ii -(∑ j e ij ) 2 )=∑ i (e ii -a i 2 )

[0086] Here, without weighting edge 52, e ii It is the ratio of the total number of edges 52 connected to each node 51 included in the cluster i to the total number of edges 52. ij is the ratio of the number of edges 52 connecting cluster i and cluster j to the number of all edges 52. i It is the ratio of the number of edges 52 connecting cluster i to other clusters to the number of all edges 52 .

[0087] In addition, when weighting the edge 52, the "number of edges 52" in the above description is replaced by the "weight of the edge 52".

[0088] The value of modularity Q varies between 0 and 1, and the closer it is to 1, the better the clustering is evaluated to be. Here, "good clustering" means that clustering is performed in a manner that more edges 52 connect nodes 51 belonging to the same cluster and fewer edges 52 connect nodes 51 belonging to different clusters.

[0089] The Girvan-Newman algorithm described above is not a method for performing clustering using modularity Q directly in the evaluation function. On the other hand, modularity Q is an indicator proposed to quantitatively evaluate the effectiveness of clustering based on the Girvan-Newman algorithm. Furthermore, by performing clustering based on the Girvan-Newman algorithm, a search is performed for the edge 52 between clusters where modularity Q becomes larger.

[0090] The Newman algorithm is a method of performing clustering while gradually increasing clusters from one node by using the amount of change in the evaluation function as the evaluation function for modularity Q. Even when the Newman algorithm is used, the device model described above can be clustered.

[0091] As described above, various methods for clustering a network represented by a graph have been proposed, represented by the Girvan-Newman algorithm. On the other hand, there has been no concept of clustering a device model of a processing device 1 that processes a fluid represented by a graph and grouping the machines 31 that constitute the processing device 1.

[0092] At this time, it is preferred that when the equipment model is created, the unique features of the processing equipment 1 can be expressed using a diagram. Figure 5 to Figure 7 To illustrate individual features and how they are expressed.

[0093] Figure 5 An example of creating a facility model taking the cost of the pipe 4 into consideration is shown.

[0094] In the processing facility 1 , the pipe 4 connecting the devices 31 to each other has various combinations of pipe diameter, wall thickness, and pipe material, and these differences are reflected in the cost per unit length of the pipe 4 .

[0095] In this case, when the pipe 4 having a high cost per unit length becomes a boundary and the devices 31 connected to the pipe 4 are divided into different device groups 3 , the pipe length of the high-cost pipe 4 may increase.

[0096] Therefore, if Figure 5 As shown in (a), the following equipment model can be created: the greater the cost per unit length of the pipe 4, the greater the weight given to the corresponding edge 52. In the figure, "unit price 1 to unit price 3" represent the relative size of the cost per unit length of the pipe 4.

[0097] exist Figure 5 In the equipment model described in (a), clustering is performed using an algorithm that can process weighted edges 52 and in which the smaller the weight, the more likely it is to be selected as the edge 52 between clusters. This makes it difficult for the equipment 31 connected to the pipe 4 with a large cost per unit length to be classified into different equipment groups 3. As an example of such an algorithm, spectral clustering can be exemplified.

[0098] Furthermore, contrary to the above example, it is not denied to adopt a method of clustering using an algorithm in which a value inversely proportional to the cost per unit length is set as a weight added to the edge 52 and the larger the weight, the more likely the edge 52 is to be selected as an edge between clusters.

[0099] In addition, using the Girvan-Newman algorithm already described, if Figure 5 As shown in (b), a facility model can be created in which the greater the cost per unit length of the pipe 4, the greater the number of edges 52 connecting the nodes 51 corresponding to the equipment 31 connected to the pipe 4.

[0100] exist Figure 5 In the equipment model described in (b), when the above-mentioned Gervan-Newman algorithm is used, since a plurality of shortest paths corresponding to the pipes 4 with a large cost per unit length are formed, the count value of the edge 52 becomes small. As a result, it is possible to make it difficult to group the equipment 31 connected to the pipes 4 with a large cost per unit length into different equipment groups 3.

[0101] Then, Figure 6 An example of creating a facility model taking into account the occupied area and evaluation value of the equipment 31 is shown.

[0102] In the processing equipment 1, the number of other machines 31 that can be arranged in the partition area 30 tends to decrease when the machine 31 occupies a large area. In addition, even when the occupied area is not large, it may not be possible to arrange a large number of other machines 31 in the partition area 30 where the machine 31 is arranged from the perspective of ensuring a safe distance or weight restrictions.

[0103] Therefore, if Figure 6 As shown in (a), the following equipment model can be created: the larger the occupied area of ​​the machine 31, or the higher the evaluation value preset for the machine 31 (the indicators required for ensuring the safety distance or weight limit mentioned above, and reducing the number of machines 31 included in the shared machine group 3), the greater the weight is. In the figure, the numerical values ​​"1, 3" added to the machine 31 represent the weight added according to the occupied area or evaluation value of the machine 31.

[0104] At this time, the edge 52 on the shortest path including the pair of nodes 51 to which the larger weight is assigned is evaluated so that the weight of the edge 52 becomes larger. Figure 6 In the example described in (a), when focusing on a pair of nodes 51 with dotted lines on the shortest path, weights of "1" and "3" are respectively added to these nodes 51. Therefore, a weight of "1×3=+3" is added to each edge 52 included in the shortest path, and the weight is counted as a count value.

[0105] On the other hand, in the above graph, when focusing on a pair of nodes 51 that are both described with a single-dot chain line on the shortest path, a weight of "1" is added to each of these nodes 51. Therefore, a weight of "1×1=+1" is added to each edge 52 included in the shortest path, and the weight is counted as a count value.

[0106] according to Figure 6 The device model described in (a) uses an algorithm that can process weighted nodes 51, and the larger the weight, the easier it is to be selected as the edge 52 between clusters. This makes it difficult to group other machines 31 into a machine group 3 that is shared with a machine 31 with a large occupied area or a machine 31 with a high evaluation value.

[0107] In addition, using the Girvan-Newman algorithm already described, if Figure 6 As shown in (b), a device model can be created that is expressed in the following manner: the larger the occupied area of ​​the machine 31 or the higher the evaluation value, the greater the number of connections of the virtual nodes 511, which are connected to a node 51 corresponding to the machine 31 via an edge 52 and are not connected to any node 51 other than the node 51.

[0108] according to Figure 6 In the device model described in (b), when the Girvan-Newman algorithm described above is used, the edge 52 on the shortest path recorded with the dotted line in the figure includes the edge 52 that is repeatedly counted multiple times according to the number of dummy nodes 511 set. On the other hand, regarding the edge 52 on the shortest path recorded with the single-point dashed line in the figure, when it is not affected by the dummy node 511, it is not repeatedly counted.

[0109] When the Gervan-Newman algorithm is executed based on the above structure, the count value of the edge 52 of the shortest path including the machine group 3 with a large occupied area or the machine group 3 with a high evaluation value becomes larger. As a result, it is possible to make it difficult for other machines 31 to be included in the machine group 3 shared with the machine 31 with a large occupied area or the machine 31 with a high evaluation value. According to the above method, the deviation of the area of ​​the division area 30 of each machine group 3 or the sum of the evaluation values ​​of the machines 31 included in the machine group 3 can be reduced.

[0110] Next, Figure 7 This is an example of creating a device model that represents the branching of the pipe 4. Figure 2 or Figure 7 As shown in (a) of FIG. 1 , among the pipes 4 installed in the processing equipment 1, there are pipes that branch into two or more pipes 4 via a branch point 40. On the other hand, in the diagram showing a general network, the representation of branching the edge 52 is not used.

[0111] Therefore, if Figure 7 As shown in (b), the following equipment model can be created: a branch node 512 is set corresponding to the branch point 40, and the branch node 512 and the node 51 corresponding to the machine 31 connected via the branched piping 4 are connected via the edge 52, thereby representing the branched piping 4.

[0112] In the above case, it is assumed that the algorithm for clustering includes the use of Figure 4A , Figure 4B The procedure for counting the edges included in the shortest path is similar to the Girvan-Newman algorithm described above. At this time, the branch node 512 is treated as not being selected as one of the two nodes 51 constituting the shortest path.

[0113] according to Figure 7 The device model described in (b) can express the branch point 40 of the pipe 4 which was difficult to express using the previous diagram.

[0114] Above, based on the use Figure 3 , Figure 5 to Figure 7 The method described above is to create a device model corresponding to the processing device 1 (process of creating a device model). Here, before clustering the created device model, clustering conditions are set (process of setting clustering conditions). The clustering conditions are constraints set when the computer performs clustering of the device model.

[0115] As the clustering condition, at least one of the number of clusters to be clustered and the range of granularity, which is the sum of weights set for nodes included in each cluster, is set.

[0116] The number of clusters is a constraint condition for setting the number of groups into which the device model is divided. When the number of clusters is set as the clustering division condition, a process of searching for edges between clusters is performed so that clusters corresponding to the set number are formed and the modularity Q becomes larger.

[0117] Granularity refers to constraints set for the cluster so that characteristics of the machine group 3, such as the number of machines 31 included in each machine group 3 or the total occupied area of ​​the machines 31 included in each machine group 3, are consistent with values ​​within a specified range.

[0118] The granularity can be calculated by adding a preset weight to the nodes 51 included in each cluster. For example, when the number of machines 31 included in the machine group 3 is consistent with the number within the specified range, a weight of "1" is set for each node 51. On the other hand, when the total occupied area of ​​the machines 31 included in each machine group 3 is consistent with the value within the specified range, a weight corresponding to the occupied area of ​​the corresponding machine 31 is set for each node 51. Figure 6 (a) Figure 7 As shown in (b), these weights can also be set when the device model is created. In addition, it may be difficult to obtain search results, but both the number of clusters to be clustered and the granularity range of the clusters can be set as clustering conditions.

[0119] After the device model is prepared and the clustering conditions are set, the device model is clustered using a computer (a process of clustering). For example, in the embodiments described below, the device model is clustered using a specially developed tool. In addition, the process can be performed using, for example, a program of the Girvan-Newman method or other clustering algorithms provided to NetworkX, a library of Python (a U.S. registered trademark of the Python Software Foundation).

[0120] As a result, yes Figure 3 The device model shown in the figure is clustered. Based on the result of the clustering, for example, Figure 8 As shown, the machines 31 constituting the processing facility 1 are grouped into a plurality of machine groups 3 .

[0121] also, Figure 8 , the result of grouping the machines 31 included in a part of the process flow of the processing equipment 1 is shown, but in fact, the machine group 3 can be grouped based on the result of creating an equipment model of the entire processing equipment 1 and performing cluster division.

[0122] Results, such as Figure 1 As schematically shown, for example, all the machines 31 constituting the processing facility 1 are grouped. Based on the result of the grouping, it is possible to advance a study of the arrangement of the machines 31 within the machine group 3 (divided area 30) or the arrangement position of the machine group 3 within the site 10. As a result, for example, it is possible to search for an arrangement with a low total usage (total cost) of the pipe 4 by adjusting the arrangement position of the machine group 3 within the site 10.

[0123] The clustering method of this embodiment has the following effects. A device model is created that represents the equipment 31 and the pipes 4 of the processing equipment 1 using a graph including nodes 51 and edges 52, and cluster division of the device model is searched for in a manner that increases modularity under pre-set conditions. As a result, quantitative evaluation can be performed based on an evaluation criterion (modularity), and the grouping of the equipment 31 constituting the processing equipment 1 can be rationally implemented.

[0124] Here, the grouping of the machines 31 does not necessarily have to be completely consistent with the clustering result of the equipment model. The clustering result is confirmed from the perspective of the technicians involved in the design and construction of the processing equipment 1. If the machine 31 included in the machine group 3 does not match the actual situation or the machine 31 is preferably included in another machine group 3, the machine group 3 including the target machine 31 can be changed.

[0125] Example

[0126] (simulation)

[0127] A. Simulation Conditions

[0128] As an example of the processing equipment 1, a natural gas liquefaction equipment (LNG equipment) whose design and configuration of the equipment 31 have been completed is modeled by PFD. Then, as a clustering condition, the number of clusters is set to "7", and clustering of the equipment model is performed by the Girvan-Newman algorithm using a specially developed tool.

[0129] The number of devices 31 constituting the LNG facility (the number of nodes 51 of the facility model) is 87. Figure 5 (a) Figure 5 (b) Figure 6 (a) Figure 6 (b) Figure 7 (a) Figure 7 (b) illustrates various adjustments to the device model.

[0130] B. Simulation Results

[0131] Fig. 9 The arrangement positions of the equipment 31 constituting the LNG facility serving as the basis of the facility model within the partition area 30 are shown. Fig.10 represents the process model with clustering implemented. In addition, Fig.10 Not included Fig. 9 The "other" machines in the

[0132] about Fig.10 The device model is divided into Fig.10 The clusters are identified by the symbols A to G added in . Fig. 9 In FIG. 3 , the machines 31 are grouped based on the result of the cluster division, and a common hatching line is added to the machines 31 included in each machine group 3 .

[0133] As in Fig. 9As shown in the hatched legend in FIG, clustering is performed in such a way that the equipment involved in the mercury removal process and a part of the equipment of the acid gas removal process (acid gas recovery unit (AGRU)) are included in the cluster of symbol A, and the remaining equipment of the AGRU is included in the cluster of symbol B. In addition, each refrigerant system of the mixed refrigerant (MR), C3 (propane) refrigerant, and N2 (nitrogen) refrigerant is clustered into different clusters (symbols E, G, and F). As for other clusters, clustering is also performed according to the content of the process implemented by the LNG equipment.

[0134] exist Fig.10 The equipment model shown does not provide any information about the function of each machine 31 or the content of natural gas processing. However, it is possible to naturally group the machines 31 according to the types of processes included in the LNG equipment or the functions of the machines 31 using only information related to the connection relationship of the machines 31 through the piping 4.

Claims

1. A clustering method, characterized in that: Including the following processes: Creating a facility model that represents a connection relationship between a plurality of machines constituting a processing facility for processing a fluid and pipes connecting the plurality of machines, using a graph that represents a connection relationship between nodes corresponding to each of the plurality of machines and edges corresponding to each of the pipes; Regarding the device model, a clustering condition is set, wherein the clustering condition is at least one of the number of clusters to be clustered and the sum of weights set on nodes included in each cluster, that is, the range of granularity; as well as The device model is clustered by searching, by a computer, for edges between clusters that satisfy the clustering conditions and have greater modularity, in, The number of clusters is a constraint condition for setting how many clusters the device model is divided into. The granularity refers to a constraint condition set for the cluster so that a characteristic of a machine group, that is, the number of the machines included in each machine group or the total occupied area of ​​the machines included in each machine group, matches a value within a predetermined range.

2. The clustering method according to claim 1, characterized in that: In the step of creating the equipment model, the equipment model is created such that the edges are weighted with a larger weight as the cost per unit length of the corresponding pipe becomes larger.

3. The clustering method according to claim 1, characterized in that: In the process of making the equipment model, the equipment model is made as follows: for the node, the larger the occupied area of ​​the corresponding machine is, or the higher the pre-set evaluation value of the machine is, the greater the weight is.

4. The clustering method according to claim 1, characterized in that: The clustering process is performed based on the Girvan-Newman algorithm.

5. The clustering method according to claim 4, characterized in that: In the step of creating the equipment model, the equipment model is created such that the number of unweighted edges connecting two nodes corresponding to two devices connected to the pipe increases as the cost per unit length of the pipe increases.

6. The clustering method according to claim 4, characterized in that: The nodes corresponding to the plurality of machines contained in the device model are weightless nodes; and In the process of making the equipment model, the following equipment model is made: the larger the occupied area of ​​the machine is or the higher the pre-set evaluation value for the machine is, the greater the number of connections of nodes, i.e., virtual nodes, which are connected to a node corresponding to the machine via an edge and not connected to nodes other than the one node.

7. The clustering method according to claim 1, characterized in that: The clustering step includes counting the edges of the shortest path between two nodes included in the device model and adding the count value; and When the processing equipment includes piping branching at a branch point, in the process of making the equipment model, the following equipment model is made: a branch node that is not selected as the two nodes when the process of adding the count value is executed and is set corresponding to the branch point, and the branch node and the node corresponding to the machine connected via the branched piping are connected via an edge, thereby representing the branched piping.

Images