CN116484006A

CN116484006A - Knowledge graph reasoning and completion-based process optimization method for process industry

Info

Publication number: CN116484006A
Application number: CN202210037347.8A
Authority: CN
Inventors: 史海波; 潘福成; 李歆; 刘泓辰; 于淼; 林百川
Original assignee: Shenyang Institute of Automation of CAS
Current assignee: Shenyang Institute of Automation of CAS
Priority date: 2022-01-13
Filing date: 2022-01-13
Publication date: 2023-07-25

Abstract

The invention discloses a process optimization method of a process industry based on knowledge graph reasoning and completion, which comprises the following steps: step one, extracting an entity set and a relation set from an optimization history, exception handling data, investigation expert knowledge and a document to construct triplet data; step two, adding priority in the triples according to the adjustment sequence of each time, and constructing a knowledge graph model with priority; step three: according to the collected real-time and historical data, the process parameters of the flow industry are optimized through knowledge graph reasoning, and associated information is output, so that timely inspection is facilitated; step four: the existing knowledge graph is complemented through the existing triplet set, and a new optimization method for the process industry is deduced. The invention can replace the control of the process in the flow industry of human expert, improves the efficiency and accuracy of the process optimization in the flow industry, and can optimize the process in the flow industry more accurately and comprehensively.

Description

Knowledge graph reasoning and completion-based process optimization method for process industry

Technical Field

The invention belongs to the field of process optimization in the process industry, and particularly relates to a process optimization method in the process industry based on knowledge graph reasoning and completion.

Background

With the rapid development of economy and science and technology in China, the manufacturing industry in China has more speaking rights in the world. But the development of the basic industry is not separated from the development of the manufacturing industry to make the manufacturing industry larger and stronger. The importance of the process industries such as metallurgical industry, chemical industry, building material industry, pharmaceutical industry and the like is becoming more prominent, and more requirements are put on products and production processes in the process industry. Therefore, many process industries have been involved, such as introducing or developing larger devices with more powerful capabilities, collecting accurate auxiliary devices, and the like. These devices are capable of obtaining large amounts of data, and therefore, by means of these data, the tuning of the process is of great importance to the modern industry.

The current optimization section of the equipment parameters in the process industry mainly depends on the working experience of experts for many years. By optimizing certain equipment, a relatively good product quality is obtained. This optimization method, which relies on expert system experience, requires extensive exploration and verification each time a new piece of equipment is added, even though there may be a large amount of missing information. Meanwhile, a great deal of manpower and time are required for culturing each expert in the flow industry, and the expert cannot judge whether the optimization is needed under the current condition in real time, because in many cases, certain indexes before a few minutes or a few hours are required to be referred to; the dilemma of expert as to whether the parameter optimization can be timely or not and whether the optimization strategy can be the optimal strategy according to the current situation is also caused. Therefore, an intelligent tuning tool facing the process of the flow industry is established, and the tuning tool has important significance for tuning and complementing the process of the flow industry.

The concept of knowledge graph is formally proposed by google corporation and 2012 to enrich the functions of search engines. Compared with the traditional expert system and knowledge engineering, the knowledge graph represents knowledge by a triplet set, each entity is a node in the graph, and each relation is an edge connecting the nodes, so that a huge graph with relation is formed. Knowledge patterns are naturally used for expressing relations, and further provide possibility for researching and connecting entities. Because of these great advantages, knowledge maps are rapidly fused and applied to various other fields, such as military, medical, public transportation, question-answering, and the like. Meanwhile, due to the participation of the fields of various industries, the techniques of expert system, natural language processing, fuzzy processing, machine learning, database, information extraction and the like further enrich the construction and application of the knowledge graph. And many communities share a large number of high-quality data sources from which more researchers analyze.

Because of the process complexity of the process industry, the current process optimization is mostly performed by an expert according to the existing experience, and indexes of a plurality of devices are ignored. In parameter optimization, other factors are often ignored; on the other hand, when the process of the process industry is optimized, many optimization rules are not found or effective experiments are not available only by virtue of years of experience of experts, so that many good optimization strategies can be lost. Therefore, the method has important practical significance and higher engineering practical value by constructing a knowledge graph facing the process of the flow industry and optimizing and complementing the process.

Disclosure of Invention

The invention provides a process optimization method based on knowledge graph reasoning and completion in the process industry, which aims at: a knowledge graph oriented to the process of the flow industry is constructed to replace the control of human experts on equipment in the flow industry and optimize and regulate the production process, so that the timeliness, efficiency and accuracy of controlling the parameter optimization of the production process are improved, and the quality and quantity of products are improved; and the completion is carried out based on the constructed knowledge graph, so that the completion of mining and supplementing of knowledge ignored or hidden by an expert is completed, and the quality of the product is improved more comprehensively.

The technical scheme adopted by the invention for achieving the purpose is as follows:

a process optimization method based on knowledge graph reasoning and completion in the process industry comprises the following steps:

taking equipment, equipment labels, production product indexes and equipment label adjustment quantity of process design in the process industry as entities to construct a triple structure of a head entity-a relation-a tail entity;

constructing a knowledge graph oriented to a process of the flow industry based on the triplet structure;

acquiring real-time data and historical data of equipment labels and production product indexes, and adjusting process parameters based on a knowledge graph to realize process optimization;

based on the existing triplet set, the knowledge graph is complemented.

The relationship is a set of relationships extracted from a process adjustment history and an exception handling history, comprising: at least one of a relationship between the equipment and the equipment label, a relationship between the equipment label and an equipment label adjustment amount, a relationship between a production product index and an equipment label adjustment amount, a relationship between the equipment label and a production product index, and a relationship between the equipment label adjustment amount and the equipment label.

The method for constructing the knowledge graph facing the process of the flow industry based on the triple structure specifically comprises the following steps:

adding priority into triples according to the set optimizing sequence of the trigger quantity in the process of the flow industry, so that each triplet (h, r, t) is formed into a corresponding quadruple (h, r, t, o), wherein h represents a head entity, r represents a trigger relation between the head entity and a tail entity, t represents a tail entity, o represents the priority of h and r pointing to t, and a plurality of quadruples form a knowledge graph facing the process of the flow industry.

The process parameter adjustment comprises the following steps:

1) Acquiring real-time data and historical data of equipment labels and production product indexes, and putting the data into four-tuple (h, r, t, o);

2) Acquiring a trigger relation r 'of the previous cycle, if r'. Noteq.r, resetting the priority of the route diffused out of all four-tuple, and executing the step 3); otherwise, executing the step 3) according to the current priority, wherein the route is a route formed by the triples executed in the knowledge graph, namely, the tail entity of the previous triplet is the head entity of the next triplet;

3) The same head entity h and the trigger relation r in the quadruple correspond to a plurality of different tail entities t, the priority corresponding to each tail entity t is different, and the tail entity t1 with the highest priority is selected as the head entity of the next step;

4) If the tail entity t1 selected in the step 3) is the equipment tag adjustment amount, optimizing the equipment parameters, namely executing the step 5); otherwise, the tail entity t1 is used as a head entity in the quadruple, and the step 2) is returned; if the tail entity t1 is the same as the current head entity h, terminating executing the line;

5) Obtaining a label corresponding to a tail entity t1, obtaining a current value v1 of the label from real-time data, and obtaining a value v2 to be regulated from equipment label regulation corresponding to the label; comparing v2 with the upper and lower threshold limits of the tag, if the value of v2 adjustment is within the threshold, writing v2 to the device; otherwise, returning to the step 2);

6) Reducing the priority of the regulated entity, waiting for the reaction time, returning to the step 1) if a new event triggering the knowledge graph exists, and otherwise, executing the step 7);

7) If the adjusted tail entity label t still has a quadruple taking the tail entity label t as a head entity, and after the equipment monitoring value and the flow production process value last for a period of time, the tail entity t is adjusted to the original value.

The knowledge graph is complemented by cycling the following formula:

wherein gamma is an edge parameter and d (h+r, t) is the distance of the correct triplet determined artificially; d, d(h+r, t') is the distance of the erroneous triplet determined by human; t (T) _batch Is a set of correct triples and incorrect triples; the error triples (h, r, t') belong to the set S "(h, r, t) if they are confirmed error triples; c is the number of times the erroneous triplet is inferred to be a triplet, and m is the distance of t' from h+r.

The invention has the following beneficial effects and advantages:

1. the invention can reduce the dependence of process optimization in the process industry on experts.

2. The invention can timely use the optimal strategy to optimize the process of the process industry

3. Optimization rules capable of supplementing flow industry expert omission or optimization rules

Drawings

FIG. 1 is a flow chart of a process optimization method based on knowledge graph reasoning process industry;

FIG. 2 is a flow chart of a process optimization method based on knowledge graph completion flow industry;

fig. 3 is a flowchart of a process optimization method based on knowledge graph reasoning and completion in the process industry, which is provided by the embodiment of the invention.

Detailed Description

The present invention will be described in further detail with reference to the accompanying drawings and examples.

The invention provides a process optimization method of a process industry based on knowledge graph reasoning and completion, which comprises the following steps:

step one, using equipment, equipment labels, production product indexes and equipment label adjustment amounts of process design in the process industry as entities of a knowledge graph, extracting relation sets from process adjustment histories and exception handling histories, and constructing a triple structure of a head entity-relation-tail entity. The entity relationship comprises a relationship between equipment and equipment labels, a relationship between equipment labels and equipment label adjustment amounts, a relationship between production product indexes and equipment label adjustment amounts, a relationship between equipment labels and production product indexes, and a relationship between equipment label adjustment amounts and equipment labels.

The specific model of the equipment, the equipment label, the process safety range and the production safety range of the equipment label, the product index in the production process, the equipment label regulating quantity and the alarm mechanism are obtained from the process regulating history and the abnormality processing history; obtaining the corresponding relation between equipment and equipment labels, the time when the equipment labels can break through the process safety range, the production intermediate quantity and the triggering range of the equipment labels for triggering and optimizing; obtaining the waiting time of different tuning strategies; the initial trigger condition of knowledge reasoning is obtained.

Step two, constructing a knowledge graph oriented to the process of the flow industry according to the triples in the knowledge base; and adding priority in the triples according to the tuning sequence of the trigger quantity in the process of the flow industry, and constructing a knowledge graph model with priority. Each triplet (h, r, t) becomes the corresponding (h, r, t, o), where o is the priority of h and l pointing to t.

The knowledge graph for the process of the flow industry consists of equipment, equipment labels, production product indexes and equipment label adjustment quantity and the connection between the equipment and the equipment label adjustment quantity. And adding the priority of triggering the tail entity to each triplet, so that the head entity can point to the tail entity with higher priority in the triplet with the same head entity and tail entity, and further triggering process optimization.

The final result of triggering tuning is in the tuning process flow, so the final purpose of reasoning is to modify the current value of the tuned equipment label into a target value, thereby tuning the quality and quantity of the production process and the final finished product. Each ring or line segment of the knowledge-graph must contain a tag for at least one device tag tuning parameter.

Step three, according to the real-time data and the historical data of the collected equipment labels and the production product indexes, the knowledge graph reasoning is oriented to parameters of the labels to be regulated in the process of industrial optimization in the process industry, the labels to be regulated and specific parameters are written into corresponding equipment, the regulation of related process parameters is locked, the reaction time is required in the production process, and then the reasoning, optimization and regulation are continued. And finally, outputting the associated information, so that the operator can check and feed back the information in time.

As shown in fig. 1, the knowledge graph for the process of the process industry is optimized, and the following steps are needed:

1. the triads (h, r, t), wherein h is a set triggering condition, and according to acquired real-time data and historical data, h and r are satisfied, a triggering knowledge graph is inferred, and priorities are included to form similar triads (h, r, t, o);

2. acquiring the last trigger relation r ', if r'. Noteq.r, resetting the priority of the route to diffuse out all triples;

3. h and r comprise a plurality of triples including the priority o1 of triplet (h, r, t 1) and the priority o2 of triplet (h, r, t 2). When o1< o2, obtaining a tail entity t1 as a trigger entity of the next step, and executing the step (4); if the tuning rule is not obtained finally, selecting a tail entity t2 to continue execution until no triplet can be executed;

4. the type of the tail entity t1 obtained from the step (3) is the equipment tag adjustment amount, the equipment parameters are optimized, and the step (5) is executed; otherwise, executing the step (2) by taking t1 as a triplet (t 1, r1, t 2) of the head entity. When t2 is the same as h, stopping executing the line to prevent a loop from occurring in the executing process;

5. obtaining a label corresponding to t1, and obtaining the current value v1 of the label and a value v2 to be adjusted; comparing v2 to upper and lower limits of the tag, writing the value of v2 adjustment to the device if the value is within a production or safe range; otherwise, repeating the adjusting step (2);

6. reducing the priority of all triples passing from the triggering entity to the executing tag adjusting entity; the purpose is to make the related triples have triggering possibility under the same condition, and prevent one triplet from being always triggered, and the rest triples are not always triggered;

the priority of the completed and confirmed triples is the lowest priority; meanwhile, as the production process in the process industry has continuity and continuity, waiting time is needed after tuning.

7. Waiting for the reaction time, and repeating the step (1) if a new trigger knowledge graph event exists;

each process optimization in the process industry requires a period of reaction time, so the last triplet of each line needs to have reaction time.

8. If the regulated tail entity label t still has a triplet taking the regulated tail entity label t as a head entity, when the equipment monitoring value and the flow production process value continue for a normal period of time, callback is carried out;

the purpose of lowering the priority is to make the related triples have triggering possibility under the same condition, thereby avoiding that one triplet is always triggered and the rest triples are not always triggered.

When the head entity and the relation correspond to a plurality of tail entities and the priorities of the tail entities are the same, the fact that the labels are written into equipment together under the condition of current process optimization is indicated, and joint tuning is carried out or downward execution is continued.

The final result of triggering tuning is to perform the adjustment process flow and optimize the label parameters. The final objective of the reasoning is thus to modify the tuned device tag from the current value to the target value, thereby tuning the production process to obtain a higher quality and a higher number of finished products.

In the knowledge graph reasoning process, real-time data and historical data of equipment labels and production product indexes are used, and meanwhile, the ascending variable quantity and the descending variable quantity of the equipment labels and the production product indexes are considered for carrying out dynamic adjustment on waiting time and an optimization strategy.

And step four, as shown in fig. 2, complementing the knowledge graph through the existing triplet set. And calculating a new process optimization rule, and manually confirming to generate a triplet. And adds priority and latency to the acknowledged triples.

And when the expert examines the completed knowledge-graph triples, the completed knowledge-graph triples are considered as wrong triples, and the completed knowledge-graph triples are put into a set S "(h, r and t) and are given greater failure weights. The completion method can be applied to a plurality of knowledge graph methods. For example, can be applied to equation (1), where the distance equation for the correct triplet (h, r, t) and the distance (h, r, t') of the wrong triplet is

Where γ is the edge parameter and d (h+r, t) is the distance of the correct triplet; d (h+r, t') is the distance of the wrong triplet; t (T) _batch Is a set of correct and incorrect triples

Because of the manual intervention of an expert, the adjustment method can further judge whether the inferred triples are correct or incorrect. For the triples which are confirmed to be wrong, a larger distance is given, so that the result meets the actual requirement of production process tuning. The modified distance formula is:

wherein the error triples (h, r, t') belong to the set S "(h, r, t) if they are confirmed error triples; c is the number of times the erroneous triplet is inferred to be a triplet and m is the length of the distance.

And finally, if d (head entity+tail entity, tail entity) is smaller than the always correct triplet, the correct triplet is considered.

Fig. 3 is a flow of process optimization in the process industry based on knowledge graph reasoning and completion, and the specific design method mainly comprises the following four steps:

1) And taking equipment, equipment labels, production product indexes and equipment label adjustment quantity of process design in the process industry as entities of the knowledge graph, extracting relation sets from process adjustment histories and exception handling histories, and constructing a triple structure of a head entity-relation-tail entity.

Taking the ore dressing process industry as an example, the obtained equipment comprises MC_QMJ, FX_CXBC and the like; the device tags are fx_802_dl_a, mc_qmj_101_kjxck_nd_re, and the like; from history tuning can be derived from tag tuning changes, e.g., MC_QMJ_101_JSLL_REAL goes from 100 to 98, MC_QMJ_101_JSLL_REAL-2 can be derived.

From the above entities, an entity set { MC_QMJ; fx_cxb; fx_802_dl_a; MC_QMJ_101_KJXCK_ND_RE; mc_qmj_101_jsll_real-2.

2) Constructing a knowledge graph oriented to the process of the flow industry according to the triples in the knowledge base; and adding priority in the triples according to the adjustment sequence of each time, and constructing a prioritized flow industry knowledge graph model. The triples { FX_SX_BCYW, FX_SX_ZFKD3-2,0}, { FX_SX_BCYW, above-normal, FX_SX_ZFKD2-1,1}, { FX_SX_BCYW, above-normal, FX_CX_CQL3-1,2}, { FX_SX_BCYW, above-normal, FX_CX_ZFKD3-1,2} and the like are obtained to form a knowledge graph of the process industry process.

3) According to the real-time data and the historical data of the equipment labels and the production product indexes, the knowledge graph reasoning is oriented to parameters needing to be adjusted when the process industry is in industrial optimization, the labels needing to be adjusted and specific parameters are written into corresponding equipment, meanwhile, the adjustment of related process parameters is locked, the reaction time is required in the production process, and then the reasoning optimization adjustment is continued. And finally, outputting the associated information, so that the operator can check and feed back the information in time.

The first trigger sequence is shown in table 1:

after the triples are triggered by the first trigger sequence in table 1, the priority of the corresponding triples is modified, and the sequence of the next trigger is shown in table 2:

TABLE 2 second trigger sequence

When triggered for the third time, two triples { FX_SX_BCYW, higher than normal, FX_CX_LD }, { FX_SX_BCYW, higher than normal, FX_CX_ZFKD3-1}, are obtained. Wherein { FX_SX_BCYW, above normal, FX_CX_ZFKD3-1} optimizes the value of FX_CX_ZFKD3-1 to the writing device; and { FX_SX_BCYW, is higher than normal, and the tail entity of FX_CX_LD } is not the tag setting, then FX_CX_LD is taken as the head entity, and the downward searching is continued.

4) And (3) pushing out new triples through the existing triplet set, and complementing the knowledge graph. M is 0 when map complement is performed for the first time; for the header entity fx_sx_bcyw and the relationship "above normal", the calculated distances are shown in table 3:

entity	Distance of
		FX_CX_LD	0.46
FX_CX_ZFKD3-1	0.58
		FX_SX_ZFKD2-1	0.64
FX_SX_ZFKD3-2	0.82
		FX_JX_CQL1+1	1.31
FX_SX_BP	0.69

Table 3 calculates the triplet distance (1 st time)

At this time, a triplet (fx_sx_bcyw, higher than normal, fx_sx_bp) is inferred. However, the expert determines through experience and process that the triplet is an erroneous triplet, the number of errors being 1. Then continuing the reasoning through the formula (2), a new calculated distance is obtained as shown in the table 4:

entity	Distance of
		FX_CX_LD	0.11
FX_CX_ZFKD3-1	0.23
		FX_SX_ZFKD2-1	0.56
FX_SX_ZFKD3-2	0.69
		FX_JX_CQL1+1	0.78
FX_SX_BP	0.54

Table 4 calculates the triplet distance (2 nd time)

It is inferred that the number of errors for the triplet (FX_SX_BCYW, above normal, FX_SX_BP) is 2. Continuing to perform the complement, the new calculated distances are shown in Table 5:

entity	Distance of
		FX_CX_LD	0.19
FX_CX_ZFKD3-1	0.37
		FX_SX_ZFKD2-1	0.49
FX_SX_ZFKD3-2	0.62
		FX_JX_CQL1+1	0.51
FX_SX_BP	0.75

Table 5 calculates the triplet distance (3 rd time)

At this point, a new triplet (fx_sx_bcyw, higher than normal, fx_jx_cql1+1) is obtained, conforming to the correct fact.

It can be seen that the obtained triplet optimization strategy meets the expectations, and the knowledge graph complement is verified to have a good effect.

Claims

1. The process optimization method for the process industry based on knowledge graph reasoning and completion is characterized by comprising the following steps of:

based on the existing triplet set, the knowledge graph is complemented.

2. The knowledge-graph-reasoning and completion-based process industry process optimization method of claim 1, wherein the relationship is a set of relationships extracted from a process adjustment history and an exception handling history, comprising: at least one of a relationship between the equipment and the equipment label, a relationship between the equipment label and an equipment label adjustment amount, a relationship between a production product index and an equipment label adjustment amount, a relationship between the equipment label and a production product index, and a relationship between the equipment label adjustment amount and the equipment label.

3. The process optimization method based on knowledge graph reasoning and completion of the process industry as claimed in claim 1, wherein the knowledge graph for the process industry is constructed based on a triplet structure, specifically:

4. The knowledge graph reasoning and completion-based process industry process optimization method according to claim 1, wherein the process parameter adjustment comprises the following steps:

5. The process optimization method based on knowledge graph reasoning and completion of the process industry according to claim 1, wherein the knowledge graph is completed by cycling the following formula:

wherein gamma is an edge parameter and d (h+r, t) is the distance of the correct triplet determined artificially; d (h+r, t') is the distance of the erroneous triplet determined by human; t (T) _batch Is the correct triplet and errorA set of false triples; the error triples (h, r, t') belong to the set S "(h, r, t) if they are confirmed error triples; c is the number of times the erroneous triplet is inferred to be a triplet, and m is the distance of t' from h+r.