CN111709193B - Automatic feature extraction method and device based on system dynamics graph - Google Patents

Abstract

The embodiment of the invention provides a method and a device for automatically extracting features based on a system dynamics diagram. The method comprises the following steps: acquiring a control logic diagram and a system dynamics diagram representing the relationship between devices in an industrial system; merging the control logic diagram and the system dynamics diagram to obtain a merged target diagram; acquiring at least one structural relationship element in the target graph, wherein the structural relationship element comprises at least one of an anchor quantity, a passage, a branch structure and a control loop; and acquiring at least one characteristic quantity for data analysis according to the at least one structural relationship element. According to the invention, the characteristic quantity which can be used for the automatic machine learning algorithm is finally obtained through the basic control logic diagram and the system dynamics diagram, the efficiency of extracting the characteristic variable in the industrial data analysis is improved, and the method has the advantages of low cost and high working efficiency.

Description

Automatic feature extraction method and device based on system dynamics graph

Technical Field

The invention relates to the technical field of industrial system feature extraction, in particular to a method and a device for automatically extracting features based on a system dynamics diagram.

Background

The automatic machine learning aims to improve the machine learning efficiency, fully utilize information such as data structure characteristics and analysis task characteristics, automatically extract characteristic variables, and perform algorithm parameter optimization and model selection. Applied to the industrial field, besides data characteristics such as time series data, the physical connection relationship of an industrial system, such as the connection relationship of pipelines (distribution of fluid flow, loss of pressure along the way, and the like), the following dynamic mechanism: material balance, energy balance, etc., as well as control logic for the system, etc., many features of the industrial field are not utilized by the general automatic machine learning (Auto-ML) algorithm.

Disclosure of Invention

The invention aims to solve the technical problem of providing a method and a device for automatically extracting characteristics based on a system dynamics diagram, so as to solve the problems that the characteristics of the existing industrial system are not completely extracted and are applied to automatic machine learning.

In order to solve the technical problems, the technical scheme of the invention is as follows:

in one aspect of the present invention, an automatic feature extraction method based on a system dynamics diagram is provided, including: acquiring a control logic diagram and a system dynamics diagram representing the relationship between devices in an industrial system; the control logic diagram represents the control logic relationship between the devices, and the system dynamics diagram represents the physical connection relationship between the devices and the position relationship of the devices in the system;

merging the control logic diagram and the system dynamics diagram to obtain a merged target diagram;

acquiring at least one structural relationship element in the target graph, wherein the structural relationship element comprises at least one of an anchor quantity, a passage, a branch structure and a control loop; wherein the anchor quantity is a node with an in-degree of 0 in the target graph;

and acquiring at least one characteristic quantity for data analysis according to the at least one structural relationship element.

Further, merging the control logic diagram and the system dynamics diagram to obtain a merged target diagram, including:

performing union operation on the nodes in the control logic diagram and the nodes in the system dynamics diagram to obtain a first diagram;

and performing union operation on the logical connection relation between the nodes in the control logic diagram and the connection relation between the nodes in the first diagram to obtain a merged target diagram.

Further, performing union operation on the nodes in the control logic diagram and the nodes in the system dynamics diagram to obtain a first diagram, including:

merging nodes with the same variable names in the control logic diagram and the system dynamics diagram into one node;

and combining the target control quantity and the basic node with the same variable name in the control logic diagram and the system dynamics diagram into a basic target node to obtain a first diagram.

Performing union operation on the logical connection relationship between the nodes in the control logic diagram and the connection relationship between the nodes in the first diagram to obtain a merged target diagram, including:

combining two connections with the same start-stop node and the same type in the control logic diagram and the first diagram into one connection;

if the control logic diagram and the first diagram have node groups, the node groups are taken as a whole, and when connection exists between the node groups and other nodes, two connections with the same starting and ending nodes and the same type are combined into one connection, and a combined target diagram is obtained; wherein the types of connections include: physical connections or logical connections between nodes.

Further, still include:

and for the newly added target nodes in the target graph, if the target nodes are obtained by calculating indexes of a plurality of original nodes, adding the original nodes to the connection between the target nodes in the target graph.

Further, obtaining at least one anchor quantity in the target map comprises:

and checking the degree of entry and the degree of exit of each node of the target graph to obtain a node with the degree of entry being 0 and the degree of exit being more than 0, and taking the node as the anchor quantity in the target graph.

Further, acquiring the longest path in the target graph without the branch and the fusion point, as the path, includes:

starting from a node with the out-degree of physical connection greater than 0 in the target graph, continuously connecting other nodes until the out-degree of physical connection of a target node is equal to 0 or a virtual node is met to form a path; the virtual nodes are sink nodes/branch nodes, wherein the sink nodes are nodes with one more input and one more output, and the branch nodes are nodes with one more input and one more output;

if the two paths are in the inclusion relationship, the longest path is reserved;

if the next node of one of the two paths is a virtual node and the next node of the other path is not a virtual node, both paths are reserved.

Further, obtaining at least one branch structure in the target graph includes:

obtaining a path which is directly connected with both a branch node and a collection node in the target graph as the branch structure; wherein, different branch paths between the same pair of branch nodes and the same pair of collection nodes are in parallel relation.

Further, acquiring at least one control loop in the target map includes:

taking a target node in the target graph as a center, and finding out all connections of the target node to form a connection set;

acquiring all nodes directly related to all connections in a connection set of the target node, wherein nodes corresponding to logical relation connections starting from the target node are controlled quantities, the set of the controlled quantities is a first node set, the set of all anchor quantities in the target graph is a second node set, and the set of other nodes is called a third node set;

circularly selecting a controlled quantity;

starting from the controlled quantity, finding out a communication path of the controlled quantity to the physical connection of each node in the third node set, adding a new connection on the path into the connection set, and adding a new node into the third node set;

taking each anchor quantity as a center, adding the anchor quantity to the connection set with the nodes in the third node set;

forming a control loop by the target node, the first node set, the second node set, the third node set and the connection set; and the physical connection which is directly connected with the nodes in the third node set but is not in the current connection set is called as the input or output connection of the current control loop.

Further, obtaining at least one characteristic quantity for data analysis based on the at least one anchor quantity, includes:

and acquiring at least one characteristic quantity in the clustering and time sequence mode segmentation of the anchor quantity according to the exogenous variable of the anchor quantity.

In another aspect of the present invention, an automatic feature extraction device based on a system dynamics diagram is provided, including:

a first acquisition module for acquiring a control logic diagram and a system dynamics diagram representing relationships between devices in an industrial system; the control logic diagram represents the control logic relationship between the devices, and the system dynamics diagram represents the physical connection relationship between the devices and the position relationship of the devices in the system;

the merging module is used for merging the control logic diagram and the system dynamics diagram to obtain a merged target diagram;

the second acquisition module is used for acquiring at least one structural relationship element in the target graph, wherein the structural relationship element comprises at least one of an anchor quantity, a passage, a branch structure and a control loop; wherein the anchor quantity is a node with an in-degree of 0 in the target graph;

and the third acquisition module is used for acquiring at least one characteristic quantity for data analysis according to the at least one structural relationship element.

The scheme of the invention at least comprises the following beneficial effects:

according to the technical scheme, the characteristic quantity which can be used for the automatic machine learning algorithm is finally obtained through the basic control logic diagram and the system dynamics diagram, the efficiency of extracting the characteristic variable in the industrial data analysis is improved, and the method has the advantages of low cost and high working efficiency.

Drawings

FIG. 1 is a step diagram of an automatic feature extraction method based on a system dynamics diagram according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a water pump speed control logic;

FIG. 3 is a system dynamics diagram;

FIG. 4 is a first view;

FIG. 5 is a target graph of the connection relationship of the node WL 1;

FIG. 6 is a target diagram after adding the correlation connection relationship of the newly added nodes WP-GP;

FIG. 7 is a target graph with its associated logical connections added to target nodes WP-GP;

FIG. 8 is a schematic illustration of anchor amount identification;

FIG. 9 is a schematic view of pathway identification;

FIG. 10 is the passageway of FIG. 9;

FIGS. 11, 12 and 13 are directed to the branched structure of FIG. 7;

FIG. 14 is a diagram of steps taken to acquire at least one control loop in a target map;

FIG. 15 is a schematic diagram of a control loop centered on WL 1;

FIG. 16 is a schematic diagram of the inner/outer control loop;

fig. 17, 18, and 19 are target diagrams in which intermediate control omission details are hidden;

fig. 20 is a device connection diagram of an automatic feature extraction device based on a system dynamics diagram in an embodiment of the present invention.

Detailed Description

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

In the invention, a system dynamics Diagram describes physical causal influence relations among different variables on the basis of a P & ID (P & ID, Piping & Instrumentation Diagram) Diagram and a system mechanism of an industrial system; the control logic diagram is a diagram describing the relationship between the target control amount (and its relationship with the original amount), the direct drive control amount (original amount). The system dynamics graph is described as a directed graph, nodes in the graph indicate that each variable is an original Node (Node), and a plurality of variables of the same point position are combined to form a variable Group Node (Group Node); a connection indicates that the relationship between two variables is directional, but does not allow the Node to have its own connection to itself (Self Loop). The physical dimension and meaning of the variable can be defined by the attribute of the variable or the name of the variable (in the embodiment of the invention, the variable is characterized by the name of the variable).

As shown in fig. 1, an embodiment of the present invention provides a method for automatically extracting features based on a system dynamics diagram, including:

s1, acquiring a control logic diagram and a system dynamics diagram representing the relationship among the devices in the industrial system; the control logic diagram represents the control logic relationship between the devices, and the system dynamics diagram represents the physical connection relationship between the devices and the position relationship of the devices in the system; here, the control logic diagram describes the relationship between the target control amount (and its relationship with the original amount), the direct drive control amount (original amount); the system dynamics diagram is used for describing physical causal influence relations among different variables on the basis of a connection relation (P & ID) diagram and a system mechanism of a system;

s2, merging the control logic diagram and the system dynamics diagram to obtain a merged target diagram;

s3, acquiring at least one structural relationship element in the target graph, wherein the structural relationship element comprises at least one of an anchor quantity, a path, a branch structure and a control loop; wherein the anchor quantity is a node with an in-degree of 0 in the target graph;

and S4, acquiring at least one characteristic quantity for data analysis according to the at least one structural relationship element.

According to the technical scheme, the characteristic quantity which can be used for the automatic machine learning algorithm is finally obtained through the basic control logic diagram and the system dynamics diagram, the efficiency of extracting the characteristic variable in the industrial data analysis is improved, and the method has the advantages of low cost and high working efficiency.

In an optional embodiment of the present invention, step S2 is implemented to merge the control logic diagram and the system dynamics diagram to obtain a merged target diagram, where the merged target diagram includes:

step S21, performing union operation on the nodes in the control logic diagram and the nodes in the system dynamics diagram to obtain a first diagram;

and step S22, performing union operation on the logical connection relationship between the nodes in the control logic diagram and the connection relationship between the nodes in the first diagram to obtain a merged target diagram.

In the embodiment of the invention, as shown in fig. 2, a schematic diagram of a water pump rotation speed control logic is shown, in the diagram, a node WP-GP represents a steam-water pressure difference, a node WL1 represents a water level of a first-loop steam generator, a node O1 represents a main valve opening degree of a first loop, a dotted line represents a control logic relationship between nodes, and a dotted line connection between nodes represents a logic connection relationship between nodes.

In this embodiment of the present invention, a system dynamics diagram is shown in FIG. 3, wherein the node WP represents the supply water pressure; WF represents the supply water flow rate; o2 and O3 indicate the main valve opening degrees of the second and third circuits; WF1 represents the water flow rate of the first circuit; WL1 denotes the first circuit steam generator water level; GF1 represents the steam flow of the first loop steam generator; GP represents steam pressure; GF represents steam flow and the solid line represents the physical connection between the nodes.

In the embodiment of the present invention, the node is also called a variable, and specifically includes: measurable original quantity (such as nodes represented by solid line ellipses in the figure), unmeasured original quantity (such as nodes represented by dotted line ellipses in the figure), target control quantity (such as nodes represented by pentagons in the figure); variables are measurable, and important but not available, quantities that can be found in data mining; a variable group is several variables that are in the same trait. The connection types include: logical relationship connections and physical connections, the connections representing relationships between variables;

in embodiments of the invention, exogenous variable/anchor quantity is a quantity that affects other variables, but is not affected by other variables; the branch is the longest communication path between the branch node and the collection node; the closed loop is connected in a logical relationship and a physical relationship to form a closed control loop.

In an alternative embodiment of the present invention, step S21 may include:

step S21, merging the nodes with the same variable name in the control logic diagram and the system dynamics diagram into one node;

and step S22, merging the target control quantity and the basic node with the same variable name in the control logic diagram and the system dynamics diagram into a basic target node to obtain a first diagram.

When the nodes in fig. 2 and fig. 3 are subjected to union operation, two nodes with the same variable name are the same, and the target control quantity and the base node with the same variable name are replaced by the target node, so that node redundancy can be eliminated. The first graph after merging is shown in fig. 4. And the WP-GP is a newly added node, and the WL1 node type is changed into a target node.

In an alternative embodiment of the present invention, step S22 includes:

step S221, merging two connections with the same start-stop node and the same type in the control logic diagram and the first diagram into one connection;

step S222, if the control logic diagram and the first diagram have node groups, the node groups are taken as a whole, and when connection exists between the node groups and other nodes, two connections with the same start-stop nodes and the same types are merged into one connection, so that a merged target diagram is obtained; wherein the types of connections include: physical connections or logical relational connections between nodes.

Further, performing union operation on the logical connection relationship between the nodes in fig. 2 and fig. 3 and the connection relationship between the nodes in the first graph to obtain a merged target graph; as shown in fig. 5, the start-stop nodes are identical and the two connections of the same type are the same connection. As shown in FIG. 6, the correlation connection relationship of the newly added nodes WP-GP is added. Such as a logical connection between WP-GP to the rotational speed node and a logical connection between power to WP-GP.

In an optional embodiment of the present invention, the method may further include:

step S223, for the newly added target node in the target graph, if the target node is calculated from the indexes of the multiple original nodes, adding the original node to the connection between the target nodes in the target graph.

In particular, for a group node, connections are allowed to the nodes inside it, as well as to the group node as a whole. When combined, the two are performed independently. And if the target node is obtained by calculating indexes of a plurality of original nodes, adding the original nodes in the target graph, wherein the original nodes are the logical connection of the target node. Fig. 7 is a target graph after adding relevant logical connections to target nodes WP-GP.

In an optional embodiment of the present invention, in step S3, the acquiring at least one anchor quantity in the target map includes:

and step S31, checking the degree of entry and the degree of exit of each node in the target graph to obtain a node with the degree of entry being 0 and the degree of exit being more than 0, and taking the node as the anchor quantity in the target graph.

Exogenous variable/anchor quantity means a quantity that affects other variables, but is not affected by other variables;

the identification method of the anchoring variable comprises the following steps: the node (basic node) with the in-degree of 0 and the out-degree of >0, if the node is the basic node in the group node, the in-degree of the group node is 0 and the out-degree is > 0. The anchor quantity is obtained by the following method: checking the in-degree and out-degree of each node in the target graph, and if the table 1 is the out-degree/in-degree corresponding to the node in the comprehensive graph, obtaining the node with the in-degree of 0 and the out-degree of more than 0 as the anchor quantity in the target graph, and the table 2 is an anchor variable identification example;

TABLE 1 outbound/inbound degrees corresponding to nodes in the Complex graph

TABLE 2 Anchorage variable identification

As shown in fig. 8, the anchor amount is power.

In an optional embodiment of the present invention, in step S3, the acquiring at least one path in the target graph includes:

step S32, obtaining the longest path in the target map without branch and fusion point as the path. The path is the longest path without branch and fusion point on the dynamic diagram of the physical system.

In an optional embodiment of the present invention, in step S32, the acquiring the longest path in the target graph where there are no branch points and no fusion points, as the path, includes:

step S321, starting from a node in the target graph, of which the out-degree of physical connection is greater than 0, continuously connecting other nodes until the out-degree of physical connection of a target node is equal to 0, or forming a path when encountering a virtual node; the virtual nodes are sink nodes/branch nodes, wherein the sink nodes are nodes with one more input and one more output, and the branch nodes are nodes with one more input and one more output;

step S322, if the two paths are in the inclusion relationship, keeping the longest path;

in step S323, if the next node of one path is a virtual node and the next node of the other path is not a virtual node, both paths are reserved.

As shown in fig. 9 and 10, the existing paths are shown as dashed boxes.

In an optional embodiment of the present invention, in step S3, the acquiring at least one branch structure in the target graph includes:

step S331, obtaining a path in the target graph, which is directly connected with both the branch node and the collection node, as the branch structure; wherein, different branch paths between the same pair of branch nodes and the same pair of collection nodes are in parallel relation.

In step S332, the branch structure is a branch structure in which two paths have a common preamble node and a common subsequent node, which indicates the longest communication path between the branch and the sink, and the path directly connected to both the branch point and the sink is the branch structure.

As shown in fig. 11 to 13, the branch structure in the target graph is shown.

As shown in fig. 14 and 15, in an alternative embodiment of the present invention, in step S3, the obtaining at least one control loop in the target map includes:

step S341, taking a target node in the target graph as a center, finding out all connections (Linkage) of the target node to form a connection set; if WL1 is the target node, all connections of WL1 constitute the connection set = { IL3, IL6, PL5, PL7 }.

Step S342, acquiring all directly related nodes connected in the connection set of the target node, where the node corresponding to the logical relationship connection starting from the target node is a controlled quantity, the set of the controlled quantity is a first node set (Control _ List), the set of all anchor quantities in the target graph is a second node set (External _ List), and the set of other nodes is referred to as a third node set (Process _ List); all nodes directly related to all connections in the connection set of WL1 are O1, WF1, GF1, power, the first set of nodes = { O1}, the second set of nodes = { power }, the third set of nodes = { WF1, GF1}, and the target node is WL 1.

Circularly selecting a controlled quantity; such as O1.

Step S343, starting from the controlled quantity, finding out the communication path of the physical connection from the controlled quantity to each node in the third node set, adding the new connection on the path into the connection set, and adding the new node into the third node set; through O1-WF 1, a connection set (Linkage _ Candiate) = { IL3, IL6, PL5, PL7, PL4}, and a third node set is unchanged, and through O1-GF 1, the connection set = { IL3, IL6, PL5, PL7, PL4, PL6}, and the third node set is unchanged.

Step S344, taking each anchor quantity as a center, adding the anchor quantity and the connection of the anchor quantity and the nodes in the third node set into the connection set; at this time, the connection set = { IL3, IL6, PL5, PL7, PL4, PL6, PL8 }.

Step S345, forming a control loop by the target node, the first node set, the second node set, the third node set, and the connection set; and the physical connection which is directly connected with the nodes in the third node set but is not in the current connection set is called as the input or output connection of the current control loop. The control loop combines logical connection and physical connection to form a closed loop control loop. For graphical representation convenience, the anchor quantity may replicate one and the same node if there are other connections. In FIG. 16, the input connection of the present control loop is PL3, and its output connection is PL 9; for convenience but without loss of generality, the intermediate control is hidden from detail, which also indicates that the control loops may be hierarchical (inner control loop, outer control loop). As shown in fig. 17, 18, and 19, the target diagram is obtained by hiding the intermediate control omission details, and in this case, the control loop is obtained by: step 1, taking the target control as a center, finding out all Linkage related to the Linkage, wherein Liknage _ Candiate = { IL1, IL2, IL4 and IL5}, as shown in FIG. 17; step 2, find out all nodes directly related to Linkage, if the Node is included in the GroupNode, the nodes are also included, at this time, Control _ List = { rotation speed }, External _ List = { power }, Process _ List = { WP, WF, GP, GF, Group _ Node1, Group _ Node3}, Objective _ List = { WP-GP }, as shown in fig. 18; step 3, starting from each controlled quantity, finding out a link _ Physical communication path from the controlled quantity to each node in the Process _ List, adding a new link on the path into the link _ Candite, adding a new node into the Process _ List, and keeping the Process _ List unchanged through O1-GroupNode 1, Liknage _ Candite = { IL1, IL2, IL4, IL5, PL1 }; by O1-GroupNode 2, link _ Candiate = { IL3, IL6, PL5, PL7, PL1, PL2, PL3, PL11, PL13, PL9, PL12, PL14, PL10}, Process _ List = { WP, WF, GP, GF, Group _ Node1, Group _ Node3, VirtualNode1, VirtualNode 2, Inner _ Loop1, Inner _ Loop2, Inner _ Loop3}, as shown in fig. 19.

Fig. 15 to 16 show a control loop with the target node WL1 as the center; FIGS. 17-19 illustrate a control loop centered on target node WP-GP;

in the drawings of the embodiments of the present invention, a small solid-line circle represents a sink node/branch node (sink node add-in-one-out, virtual node, or branch node add-in-one), a solid-line arrow represents a Physical relationship stream (link _ Physical), a dashed-line arrow represents an Information/control stream (link _ Information), except for the sink node/branch node, other types of nodes allow add-in-multiple-out, a rectangle with a camber angle represents a variable group (group node) of one point, and a group node is multiple variables (or nodes) representing the same Physical basic point, so that, for modeling convenience only, the link can directly go to the group node, represents all basic nodes that can be directly connected to the inside of the group node, and also can go to a certain basic node inside the group node. The variables are labeled as in Table 3:

variables of	Description of the invention
		O1	Opening degree of main valve of first loop
WF1	Water flow of the first circuit
		GF1	Steam flow of first loop steam generator
WL1	Water level of first loop steam generator
		GP	Steam pressure
WP	Pressure of water supply
		WP-GP	Pressure difference between steam and water
GF	Flow rate of steam
		WF	Flow rate of water supply
O2、O3	Opening of main valve of second and third circuits

In an optional embodiment of the present invention, in step S4, obtaining at least one feature quantity for data analysis according to the at least one anchor quantity includes:

step S41, according to the exogenous variable of the at least one anchor quantity, at least one characteristic quantity in the clustering and time sequence mode segmentation of the anchor quantity is obtained.

In an optional embodiment of the present invention, in step S4, obtaining at least one feature quantity for data analysis according to the at least one path includes:

step S42, according to the measurement quantities of the same type of measurement nodes at different positions of the path, at least one of the following characteristic quantities is obtained: the difference value of the measured quantity, the mean value of the measured quantity, and the absolute deviation or the relative deviation of the single-point measured quantity and the mean value; and/or

And step S43, acquiring a functional relation between the opening and the flow according to the opening and the flow of the passage.

In an optional embodiment of the present invention, in step S4, the obtaining, according to the at least one branch structure, at least one feature quantity for data analysis includes at least one of:

step S44, obtaining distribution ratio characteristic quantity according to the branch flow distribution of the at least one branch structure;

step S45, obtaining the dimension mean value of a plurality of branch structures and the absolute deviation or the relative deviation of the single point and the dimension mean value according to the same dimension comparison of different branch structures of the at least one branch structure;

and step S46, obtaining the balance relation and/or pressure drop of the flow of the plurality of branch structures according to the same dimension comparison of the branch and the total branch in the at least one branch structure.

In an optional embodiment of the present invention, in step S4, the obtaining at least one feature quantity for data analysis according to the at least one control loop includes:

and step S47, acquiring the time sequence mode of the control loop according to the relation between the target node of the control loop and the control effect or the relation between the control node and the control action.

In the embodiment of the invention, the control logic diagram and the system dynamics diagram are merged to obtain a merged target diagram; acquiring at least one structural relationship element in the target graph, wherein the structural relationship element comprises at least one of an anchor quantity, a passage, a branch structure and a control loop; wherein the anchor quantity is a node with an in-degree of 0 in the target graph; acquiring at least one characteristic quantity for data analysis according to the at least one structural relationship element; when the characteristic quantities are used for data analysis, the machine learning efficiency can be improved.

As shown in fig. 20, an embodiment of the present invention provides an automatic feature extraction device based on a system dynamics diagram, including:

a first acquisition module for acquiring a control logic diagram and a system dynamics diagram representing relationships between devices in an industrial system; the control logic diagram represents the control logic relationship between the devices, and the system dynamics diagram represents the physical connection relationship between the devices and the position relationship of the devices in the system;

the merging module is used for merging the control logic diagram and the system dynamics diagram to obtain a merged target diagram;

the second acquisition module is used for acquiring at least one structural relationship element in the target graph, wherein the structural relationship element comprises at least one of an anchor quantity, a passage, a branch structure and a control loop; wherein the anchor quantity is a node with an in-degree of 0 in the target graph;

and the third acquisition module is used for acquiring at least one characteristic quantity for data analysis according to the at least one structural relationship element.

According to the technical scheme, the characteristic quantity which can be used for the automatic machine learning algorithm is finally obtained through the basic control logic diagram and the system dynamics diagram, the efficiency of extracting the characteristic variable in the industrial data analysis is improved, and the method has the advantages of low cost and high working efficiency.

In an optional embodiment of the present invention, the merging module is specifically configured to:

performing union operation on the nodes in the control logic diagram and the nodes in the system dynamics diagram to obtain a first diagram;

and performing union operation on the logical connection relation between the nodes in the control logic diagram and the connection relation between the nodes in the first diagram to obtain a merged target diagram.

In an optional embodiment of the present invention, the merging module is specifically configured to:

merging nodes with the same variable names in the control logic diagram and the system dynamics diagram into one node;

and combining the target control quantity and the basic node with the same variable name in the control logic diagram and the system dynamics diagram into a basic target node to obtain a first diagram.

In an optional embodiment of the present invention, the merging module is specifically configured to:

combining two connections with the same start-stop node and the same type in the control logic diagram and the first diagram into one connection;

if the control logic diagram and the first diagram have node groups, the node groups are taken as a whole, and when connection exists between the node groups and other nodes, two connections with the same starting and ending nodes and the same type are combined into one connection, and a combined target diagram is obtained; wherein the types of connections include: physical connections or logical relational connections between nodes.

In an optional embodiment of the present invention, the merging module is further configured to:

and for the newly added target nodes in the target graph, if the target nodes are obtained by calculating indexes of a plurality of original nodes, adding the original nodes to the connection between the target nodes in the target graph.

In particular, for a group node, connections are allowed to the nodes inside it, as well as to the group node as a whole. When combined, the two are performed independently. And if the target node is obtained by calculating indexes of a plurality of original nodes, adding the original nodes in the target graph, wherein the original nodes are the logical connection of the target node.

In an optional embodiment of the present invention, the second obtaining module is specifically configured to:

and checking the degree of entry and the degree of exit of each node in the target graph to obtain a node with the degree of entry being 0 and the degree of exit being more than 0, and taking the node as the anchor quantity in the target graph.

In an optional embodiment of the present invention, the second obtaining module is specifically configured to:

and acquiring the longest path without branch and fusion points in the target graph as the path. The path is the longest path without branch and fusion point on the dynamic diagram of the physical system.

In an optional embodiment of the present invention, the second obtaining module is specifically configured to:

starting from a node with the out-degree of physical connection greater than 0 in the target graph, continuously connecting other nodes until the out-degree of physical connection of a target node is equal to 0 or a virtual node is met to form a path; the virtual nodes are collection points/branch points, wherein the collection points are nodes which are input and output in addition, and the branch points are nodes which are input and output in addition;

if the two paths are in the inclusion relationship, the longest path is reserved;

if the next node of one of the two paths is a virtual node and the next node of the other path is not a virtual node, both paths are reserved.

In an optional embodiment of the present invention, the second obtaining module is specifically configured to:

obtaining a path which is directly connected with both a branch node and a collection node in the target graph as the branch structure; wherein, different branch paths between the same pair of branch nodes and the same pair of collection nodes are in parallel relation.

The branch structure is that two paths have a common preamble node and common subsequent node, and represents the longest communication path between the branch node and the sink node, and the path which is directly connected with the branch node and the sink node is the branch structure.

In an optional embodiment of the present invention, the second obtaining module is specifically configured to:

taking a target node in the target graph as a center, finding out all connections (links) of the target node to form a connection set;

acquiring all directly related nodes connected in a connection set of the target node, wherein the nodes corresponding to the logical relationship connection starting from the target node are controlled quantities, the set of the controlled quantities is a first node set (Control _ List), the set of all anchor quantities in the target graph is a second node set (External _ List), and the set of other nodes is called a third node set (Process _ List);

circularly selecting a controlled quantity;

starting from the controlled quantity, finding out a communication path of the controlled quantity to the physical connection of each node in the third node set, adding a new connection on the path into the connection set, and adding a new node into the third node set;

taking each anchor quantity as a center, adding the anchor quantity to the connection set with the nodes in the third node set;

forming a control loop by the target node, the first node set, the second node set, the third node set and the connection set; and the physical connection which is directly connected with the nodes in the third node set but is not in the current connection set is called as the input or output connection of the current control loop. The control loop combines logical connection and physical connection to form a closed loop control loop.

In an optional embodiment of the present invention, the third obtaining module is specifically configured to:

and acquiring at least one characteristic quantity in the clustering and time sequence mode segmentation of the anchor quantity according to the exogenous variable of the anchor quantity.

In an optional embodiment of the present invention, the third obtaining module is specifically configured to:

according to the measurement quantities of the same type of measurement nodes at different positions of the path, at least one of the following characteristic quantities is obtained: the difference value of the measured quantity, the mean value of the measured quantity, and the absolute deviation or the relative deviation of the single-point measured quantity and the mean value; and/or

And acquiring a functional relation between the opening degree and the flow rate according to the opening degree and the flow rate of the passage.

In an optional embodiment of the present invention, the third obtaining module is specifically configured to:

acquiring distribution ratio characteristic quantity according to the branch flow distribution of the at least one branch structure;

obtaining a dimension mean value of a plurality of branch structures and absolute deviation or relative deviation of a single point and the dimension mean value according to the same dimension comparison of different branch structures of the at least one branch structure;

and obtaining the balance relation and/or the pressure drop of the flow of the plurality of branch structures according to the isometry comparison of the branch and the total branch in the at least one branch structure.

In an optional embodiment of the present invention, the third obtaining module is specifically configured to:

and acquiring the time sequence mode of the control loop according to the relation between the target node of the control loop and the control effect or the relation between the control node and the control action.

In the embodiment of the invention, the control logic diagram and the system dynamics diagram are merged to obtain a merged target diagram; acquiring at least one structural relationship element in the target graph, wherein the structural relationship element comprises at least one of an anchor quantity, a passage, a branch structure and a control loop; wherein the anchor quantity is a node with an in-degree of 0 in the target graph; acquiring at least one characteristic quantity for data analysis according to the at least one structural relationship element; when the characteristic quantities are used for data analysis, the machine learning efficiency can be improved.

It should be noted that the apparatus is an apparatus corresponding to the method described in fig. 1, and all the implementations of the illustrated method are applicable to the embodiment of the apparatus, and the same technical effects can be achieved.

Obtaining at least one characteristic quantity for data analysis according to the at least one structural relationship element, where table 4 is a rule table for obtaining characteristic quantities for different structural relationship elements, and specifically includes: acquiring at least one characteristic quantity in clustering and time sequence mode segmentation of the anchor quantity according to the exogenous variable of the anchor quantity; or according to the measurement quantities of the same type of measurement nodes at different positions of the path, obtaining at least one of the following characteristic quantities: the difference value of the measured quantity, the mean value of the measured quantity, and the absolute deviation or the relative deviation of the single-point measured quantity and the mean value; and/or acquiring a functional relation between the opening degree and the flow rate according to the opening degree and the flow rate of the passage; or obtaining distribution ratio characteristic quantity according to the branch flow distribution of the at least one branch structure; obtaining a dimension mean value of a plurality of branch structures and absolute deviation or relative deviation of a single point and the dimension mean value according to the same dimension comparison of different branch structures of the at least one branch structure; obtaining the balance relation and/or pressure drop of the flow of a plurality of branch structures according to the isometry comparison of the branches and the total branches in the at least one branch structure; or acquiring the time sequence mode of the control loop according to the relation between the target node of the control loop and the control effect or the relation between the control node and the control action.

Table 4 rule table for obtaining feature quantities of different structural relationship elements

While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined in the appended claims.

Claims (8)

1. A method for automatically extracting features based on a system dynamics graph is characterized by comprising the following steps:

acquiring a control logic diagram and a system dynamics diagram representing the relationship between devices in an industrial system; the control logic diagram represents the control logic relationship between the devices, and the system dynamics diagram represents the physical connection relationship between the devices and the position relationship of the devices in the system;

merging the control logic diagram and the system dynamics diagram to obtain a merged target diagram;

acquiring at least one structural relationship element in the target graph, wherein the structural relationship element comprises at least one of an anchor quantity, a passage, a branch structure and a control loop; wherein the anchor quantity is a node with an in-degree of 0 in the target graph;

acquiring at least one characteristic quantity for data analysis according to the at least one structural relationship element;

the method for obtaining the combined target map by combining the control logic map and the system dynamics map comprises the following steps:

merging nodes with the same variable names in the control logic diagram and the system dynamics diagram into one node;

combining the target control quantity with the same variable name in the control logic diagram and the system dynamics diagram with a basic node to obtain a basic target node, and obtaining a first diagram;

combining two connections with the same start-stop node and the same type in the control logic diagram and the first diagram into one connection;

if the control logic diagram and the first diagram have node groups, the node groups are taken as a whole, and when connection exists between the node groups and other nodes, two connections with the same starting and ending nodes and the same type are combined into one connection, and a combined target diagram is obtained; wherein the types of connections include: physical connections or logical connections between nodes.

2. The method for automatically extracting features based on the system dynamics graph according to claim 1, further comprising:

and for the newly added target nodes in the target graph, if the target nodes are obtained by calculating indexes of a plurality of original nodes, adding the original nodes to the connection between the target nodes in the target graph.

3. The method for automatically extracting features based on the system dynamics graph according to claim 2, wherein the obtaining of at least one anchor quantity in the target graph comprises:

and checking the degree of entry and the degree of exit of each node of the target graph to obtain a node with the degree of entry being 0 and the degree of exit being more than 0, and taking the node as the anchor quantity in the target graph.

4. The method according to claim 3, wherein the step of obtaining the longest path without branch and fusion points in the target map as the path comprises:

starting from a node with the out-degree of physical connection greater than 0 in the target graph, continuously connecting other nodes until the out-degree of physical connection of a target node is equal to 0 or a virtual node is met to form a path; the virtual nodes are sink nodes/branch nodes, wherein the sink nodes are nodes with one more input and one more output, and the branch nodes are nodes with one more input and one more output;

if the two paths are in the inclusion relationship, the longest path is reserved;

if the next node of one of the two paths is a virtual node and the next node of the other path is not a virtual node, both paths are reserved.

5. The method according to claim 4, wherein the obtaining of at least one branch structure in the target map comprises:

obtaining a path which is directly connected with both a branch node and a collection node in the target graph as the branch structure; wherein, different branch paths between the same pair of branch nodes and the same pair of collection nodes are in parallel relation.

6. The method for automatically extracting features based on the system dynamics graph according to claim 5, wherein the obtaining of at least one control loop in the target graph comprises:

taking a target node in the target graph as a center, and finding out all connections of the target node to form a connection set;

acquiring all nodes directly related to all connections in a connection set of the target node, wherein nodes corresponding to logical relation connections starting from the target node are controlled quantities, the set of the controlled quantities is a first node set, the set of all anchor quantities in the target graph is a second node set, and the set of other nodes is called a third node set;

circularly selecting a controlled quantity;

starting from the controlled quantity, finding out a communication path of the controlled quantity to the physical connection of each node in the third node set, adding a new connection on the path into the connection set, and adding a new node into the third node set;

taking each anchor quantity as a center, adding the anchor quantity and the connection of the anchor quantity and the nodes in the third node set into the connection set;

forming a control loop by the target node, the first node set, the second node set, the third node set and the connection set; and the physical connection which is directly connected with the nodes in the third node set but is not in the current connection set is called as the input or output connection of the current control loop.

7. The method according to claim 6, wherein obtaining at least one feature quantity for data analysis based on the at least one anchor quantity comprises:

and acquiring at least one characteristic quantity in the clustering and time sequence mode segmentation of the anchor quantity according to the exogenous variable of the anchor quantity.

8. An automatic feature extraction device based on a system dynamics diagram is characterized by comprising:

a first acquisition module for acquiring a control logic diagram and a system dynamics diagram representing relationships between devices in an industrial system; the control logic diagram represents the control logic relationship between the devices, and the system dynamics diagram represents the physical connection relationship between the devices and the position relationship of the devices in the system;

the merging module is used for merging the control logic diagram and the system dynamics diagram to obtain a merged target diagram;

the second acquisition module is used for acquiring at least one structural relationship element in the target graph, wherein the structural relationship element comprises at least one of an anchor quantity, a passage, a branch structure and a control loop; wherein the anchor quantity is a node with an in-degree of 0 in the target graph;

a third obtaining module, configured to obtain at least one feature quantity for data analysis according to the at least one structural relationship element;

the method for obtaining the combined target map by combining the control logic map and the system dynamics map comprises the following steps:

merging nodes with the same variable names in the control logic diagram and the system dynamics diagram into one node;

combining the target control quantity with the same variable name in the control logic diagram and the system dynamics diagram with a basic node to obtain a basic target node, and obtaining a first diagram;

combining two connections with the same start-stop node and the same type in the control logic diagram and the first diagram into one connection;

if the control logic diagram and the first diagram have node groups, the node groups are taken as a whole, and when connection exists between the node groups and other nodes, two connections with the same starting and ending nodes and the same type are combined into one connection, and a combined target diagram is obtained; wherein the types of connections include: physical connections or logical connections between nodes.

Images