A Process Optimization Method for the Industry Based on Knowledge Graph Reasoning and Completion

Abstract

This invention discloses a process optimization method for the industry based on knowledge graph reasoning and completion. The steps are as follows: Step 1: Extract entity sets and relation sets from optimization history, anomaly handling data, expert knowledge, and documents to construct triplet data; Step 2: Add priorities to the triples according to the order of each adjustment to construct a prioritized knowledge graph model; Step 3: Optimize process parameters for the industry through knowledge graph reasoning based on collected real-time and historical data, and output related information for timely review; Step 4: Complete the existing knowledge graph using the existing triplet set to deduce new optimization methods for the industry. This invention can replace human experts in controlling processes in the industry, improving the efficiency and accuracy of process optimization, and enabling more precise and comprehensive optimization of process processes.

Description

Technical Field

[0001] This invention belongs to the field of process optimization in the process industry, specifically a process optimization method based on knowledge graph reasoning and completion. Background Technology

[0002] With the rapid development of my country's economy and technology, my country's manufacturing industry has gained greater influence in the world. However, to make the manufacturing industry bigger and stronger, the development of basic industries is indispensable. The importance of process industries such as metallurgy, chemicals, building materials, and pharmaceuticals is becoming increasingly prominent, placing higher demands on their products and production processes. Therefore, process industries have done a lot, such as introducing or developing larger, more powerful equipment and more precise auxiliary equipment. This equipment can acquire a large amount of data, which is crucial for optimizing processes and is of great significance to modern industry.

[0003] Currently, equipment parameter optimization in the process industry primarily relies on the years of experience of experts. Optimizing certain equipment achieves a relatively better product quality. This expert-driven optimization method requires large-scale exploration and verification each time a new piece of equipment is added, and even then, significant information may be missed. Furthermore, training a process industry expert requires substantial manpower and time, and experts cannot make real-time judgments about whether optimization is necessary in the current situation, as they often need to refer to indicators from minutes or hours ago. This creates a dilemma for experts regarding the timeliness of parameter optimization and whether the optimization strategy is optimal. Therefore, establishing an intelligent process industry optimization tool to fine-tune and complete process technologies is of great significance to the process industry.

[0004] The concept of knowledge graphs was formally proposed by Google in 2012 to enrich the functionality of its search engine. Compared to traditional expert systems and knowledge engineering, knowledge graphs represent knowledge using sets of triples, where each entity is a node in the graph and each relation is an edge connecting the nodes, thus forming a vast, relational graph. Knowledge graphs are inherently designed to express relationships and further provide the possibility of studying and connecting entities. Due to these significant advantages, knowledge graphs have been rapidly integrated and applied to various other fields, such as military, healthcare, public transportation, and question answering. Simultaneously, the involvement of these diverse industries has further enriched the construction and application of knowledge graphs through technologies such as expert systems, natural language processing, fuzzy logic, machine learning, databases, and information extraction. Moreover, many communities share a wealth of high-quality data sources, allowing more researchers to analyze them.

[0005] Due to the complexity of processes in the industry, current process optimization largely relies on experts optimizing parameters based on existing experience, often overlooking many equipment specifications. Parameter optimization frequently ignores other factors; furthermore, in process industry optimization, relying solely on years of expert experience means many optimization rules remain undiscovered or lack effective experimental verification, thus missing many valuable optimization strategies. Therefore, constructing a knowledge graph for process industry processes and optimizing and supplementing these processes has significant practical implications and high engineering value. Summary of the Invention

[0006] This invention proposes a process optimization method for the process industry based on knowledge graph reasoning and completion. Its purpose is to: construct a knowledge graph for process industries, replacing human experts in controlling equipment within these industries, optimizing and adjusting the production process, thereby improving the timeliness, efficiency, and accuracy of process parameter optimization, and ultimately enhancing product quality and quantity; and to complete the knowledge graph by uncovering and supplementing knowledge that experts have overlooked or hidden, further improving product quality more comprehensively.

[0007] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0008] A process optimization method for the industry based on knowledge graph reasoning and completion includes the following steps:

[0009] In the process industry, the equipment, equipment labels, production product indicators, and equipment label adjustment quantities in process design are used as entities to construct a triplet structure of head entity-relationship-tail entity;

[0010] Based on the triplet structure, a knowledge graph for process industry technologies is constructed.

[0011] Acquire real-time and historical data of equipment tags and production product indicators, and adjust process parameters based on knowledge graphs to achieve process optimization;

[0012] Based on the existing set of triples, complete the knowledge graph.

[0013] The relationships are a set of relationships extracted from process adjustment history and anomaly handling history, including at least one of the following: relationship between equipment and equipment tag, relationship between equipment tag and equipment tag adjustment amount, relationship between production product index and equipment tag adjustment amount, relationship between equipment tag and production product index, and relationship between equipment tag adjustment amount and equipment tag.

[0014] The construction of a knowledge graph based on a triplet structure for process industries specifically involves:

[0015] Priorities are added to triples according to the optimization order of trigger quantities in the process industry, so that each triple (h,r,t) becomes a corresponding quad (h,r,t,o), where h represents the head entity, r represents the trigger relationship between the head entity and the tail entity, t represents the tail entity, and o represents the priority of h and r pointing to t. Several quads constitute a knowledge graph for process industry processes.

[0016] The adjustment of process parameters includes the following steps:

[0017] 1) Obtain real-time and historical data of equipment tags and production product indicators, and put the data into a quadruple (h,r,t,o);

[0018] 2) Obtain the trigger relationship r' of the previous loop. If r' ≠ r, reset the priority of all quadruplets that the route spreads out and execute step 3); otherwise, execute step 3) according to the current priority. The route is a route composed of triples that have been executed in the knowledge graph, that is, the tail entity of the previous triple is the head entity of the next triple.

[0019] 3) In the quadruple, the same head entity h and triggering relationship r correspond to multiple different tail entities t. Each tail entity t has a different priority. The tail entity t1 with the highest priority is selected as the head entity for the next step.

[0020] 4) If the tail entity t1 selected in step 3) is the device tag adjustment amount, then optimize the device parameters, i.e., execute step 5); otherwise, use the tail entity t1 as the head entity in the quadruple and return to step 2); if the tail entity t1 is the same as the current head entity h, terminate the execution of this line.

[0021] 5) Obtain the tag corresponding to the tail entity t1, and obtain the current value v1 of the tag from the real-time data. Obtain the value v2 to be adjusted from the device tag adjustment amount corresponding to the tag. Compare v2 with the upper and lower limits of the threshold of the tag. If the value of v2 is within the threshold, write v2 to the device; otherwise, return to step 2).

[0022] 6) Lower the priority of the adjusted entity and wait for the reaction time. If there is a new event that triggers the knowledge graph, return to step 1); otherwise, execute step 7.

[0023] 7) If the adjusted tail entity label t still has a quadruple with it as the head entity, after the equipment monitoring value and process production value have been maintained for a period of time, the tail entity t will be adjusted back to its original value.

[0024] Complete the knowledge graph by iterating through the following formula:

[0025]

[0026] Where γ is the marginal parameter, d(h+r,t) is the distance to the correctly judged triplet by the human; d(h+r,t') is the distance to the incorrectly judged triplet by the human; T batch It is the set of correct triples and incorrect triples; if an incorrect triple (h,r,t') is a confirmed incorrect triple, then it also belongs to the set S"(h,r,t); c is the number of times the incorrect triple is inferred to be a triple, and m is the distance between t' and h+r.

[0027] The present invention has the following beneficial effects and advantages:

[0028] 1. This invention can reduce the reliance of process industries on experts for process optimization.

[0029] 2. This invention enables timely optimization of processes in the process industry using the optimal strategy.

[0030] 3. It can supplement or optimize rules that have been overlooked by industry experts. Attached Figure Description

[0031] Figure 1 It is a flowchart of an industry process optimization method based on knowledge graph reasoning.

[0032] Figure 2 It is a flowchart of an industry process optimization method based on knowledge graph completion.

[0033] Figure 3 This is a flowchart of a process optimization method for industry processes based on knowledge graph reasoning and completion provided by the present invention. Detailed Implementation

[0034] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0035] This invention proposes a process optimization method for the industry based on knowledge graph reasoning and completion, including:

[0036] Step 1: Treat the equipment, equipment tags, production product indicators, and equipment tag adjustment quantities in the process industry as entities in a knowledge graph. Extract relationship sets from the process adjustment history and anomaly handling history to construct a triplet structure of head entity-relationship-tail entity. The entity relationships include the relationships between equipment and equipment tags, equipment tags and equipment tag adjustment quantities, production product indicators and equipment tag adjustment quantities, equipment tags and production product indicators, and equipment tag adjustment quantities and equipment tags.

[0037] The system obtains the specific equipment model, equipment tag, process safety range and production safety range of the equipment tag, product indicators during production, equipment tag adjustment amount, and alarm mechanism from the process adjustment history and anomaly handling history; it also obtains the correspondence between equipment and equipment tag, the timing when the equipment tag can exceed the process safety range, and the trigger range for optimization triggered by the production intermediate quantity and equipment tag; it obtains the waiting time for different optimization strategies; and it obtains the starting trigger conditions for knowledge reasoning.

[0038] Step 2: Construct a knowledge graph for process industry processes based on the triples in the knowledge base; add priorities to the triples according to the optimization order of trigger quantities in the process industry processes, and construct a knowledge graph model with priorities. Each triple (h,r,t) becomes the corresponding (h,r,t,o), where o is the priority of h and l pointing to t.

[0039] The knowledge graph for process industries consists of equipment, equipment tags, production product indicators, equipment tag adjustment quantities, and the relationships between them. Each triple is assigned a priority to trigger a tail entity, ensuring that in triples with the same head and tail entities, the head entity points to the higher-priority tail entity, thereby triggering process optimization.

[0040] The ultimate result of triggering optimization is always optimizing the process flow. Therefore, the ultimate goal of inference is to modify the current value of the equipment label to the target value, thereby optimizing the production process and the quality and quantity of the final product. Each ring or segment of the knowledge graph must contain at least one label for the equipment label adjustment parameter.

[0041] Step 3: Based on real-time and historical data of collected equipment tags and production product indicators, the knowledge graph inference, targeting industrial optimization in the process industry, requires adjusting the parameters of the tags that need adjustment. The tags and specific parameters to be adjusted are written into the corresponding equipment, and the adjustment of relevant process parameters is locked, allowing the production process sufficient reaction time. Then, inference and optimization adjustments continue. Finally, the associated information is output for operators to check and provide feedback promptly.

[0042] like Figure 1 As shown, knowledge graph optimization for process industries requires the following steps:

[0043] 1. Triple (h,r,t) where h is the set trigger condition. Based on the real-time and historical data collected, if h and r are satisfied, the knowledge graph is triggered to perform reasoning. It also includes priority to form a similar quadple (h,r,t,o).

[0044] 2. Obtain the previous trigger relationship r'. If r' ≠ r, reset the priority of all triples propagated from this route.

[0045] 3. h and r encompass multiple triples, including the priority o1 of triple (h, r, t1) and the priority o2 of triple (h, r, t2). When o1 < o2, obtain the tail entity t1 as the trigger entity for the next step and execute step (4); if no optimization rule is finally obtained, select the tail entity t2 and continue to execute until no triple can be executed;

[0046] 4. If the type of the tail entity t1 obtained from step (3) is the device label adjustment amount, optimize the device parameters and execute step (5); otherwise, execute step (2) for the triple (t1, r1, t2) with t1 as the head entity. When t2 is the same as h, abort the execution of this line to prevent loops during the execution process;

[0047] 5. Obtain the label corresponding to t1 and the current value v1 of this label, and the value v2 to be adjusted to; compare v2 with the upper and lower limits of this label. If the adjusted value of v2 is within the production or safety range, write this value to the device; otherwise, repeat the adjustment step (2);

[0048] 6. Lower the priorities of all triples passed from the trigger entity to the execution label adjustment entity; the purpose is to make all relevant triples have the possibility of being triggered under the same conditions, preventing a certain triple from being triggered continuously while the remaining triples are never triggered;

[0049] Set the priority of the completed and confirmed triples to the lowest priority; at the same time, due to the continuity and persistence of the production process in the process industry, a waiting time is required after optimization.

[0050] 7. Wait for the reaction time. If there is a new trigger knowledge graph event, repeat step (1);

[0051] Each process optimization in the process industry requires a certain reaction time, so each triple at the end of each line needs to have a reaction time.

[0052] 8. If there are still triples with the adjusted tail entity label t as the head entity, when the device monitoring value and the process production process value remain normal for a certain period of time, perform a callback;

[0053] The purpose of lowering the priority is to make all relevant triples have the possibility of being triggered under the same conditions, preventing a certain triple from being triggered continuously while the remaining triples are never triggered.

[0054] When the head entity and the relationship correspond to multiple tail entities and the priorities of these tail entities are the same, it means that under the current process optimization situation, these several labels need to be jointly written into the device for joint optimization or continue to execute downward.

[0055] The ultimate goal of triggering optimization is to adjust the process flow and optimize label parameters. Therefore, the ultimate purpose of inference is to modify the equipment label from its current value to the target value, thereby optimizing the production process and obtaining higher quality and greater quantity of finished products.

[0056] The knowledge graph reasoning process not only uses real-time and historical data of equipment tags and production product indicators, but also takes into account their rising and falling changes, and dynamically adjusts the waiting time and optimization strategy.

[0057] like Figure 2 As shown, step four involves completing the knowledge graph using the existing set of triples. New process optimization rules are calculated, manually confirmed, and triples are generated. Priority and waiting time are then assigned to the confirmed triples.

[0058] When the completed knowledge graph triples are reviewed by experts and deemed incorrect, they are added to the set S(h,r,t) and given a higher failure weight. This completion method can be applied to many knowledge graph methods. For example, it can be applied to formula (1), where the distance formula between the correct triple (h,r,t) and the incorrect triple (h,r,t') is:

[0059]

[0060] Where γ is the marginal parameter, d(h+r,t) is the distance to the correct triplet, d(h+r,t') is the distance to the incorrect triplet, and T batch It is the set of correct triples and incorrect triples.

[0061] Because this adjustment method involves expert intervention, it is more accurate to determine whether the derived triples are correct or incorrect. For triples confirmed to be incorrect, a larger distance is assigned, making the results more aligned with the actual needs of production process optimization. The modified distance formula is:

[0062]

[0063] Where the erroneous triple (h,r,t') is a confirmed erroneous triple and belongs to the set S"(h,r,t); c is the number of times the erroneous triple is inferred to be a triple, and m is the distance from the erroneous triple.

[0064] If the final result d(head entity + tail entity, tail entity) is less than a triplet that has always been correct, then it is considered a correct triplet.

[0065] Figure 3 The process optimization method for industry processes based on knowledge graph reasoning and completion provided by this invention mainly includes the following four steps:

[0066] 1) Take the equipment, equipment labels, production product indicators, and equipment label adjustment quantities in the process industry as entities in the knowledge graph, extract the relation set from the process adjustment history and anomaly handling history, and construct a triplet structure of head entity-relationship-tail entity.

[0067] Taking the mineral processing industry as an example, the equipment obtained includes MC_QMJ, FX_CXBC, etc.; the equipment tags include FX_802_DL_A, MC_QMJ_101_KJXCK_ND_RE, etc.; historical optimization can be obtained from the tag adjustment changes, such as MC_QMJ_101_JSLL_REAL changing from 100 to 98, which can yield MC_QMJ_101_JSLL_REAL-2.

[0068] From the above entities, the entity set {MC_QMJ; FX_CXB; FX_802_DL_A; MC_QMJ_101_KJXCK_ND_RE; MC_QMJ_101_JSLL_REAL-2......} is formed.

[0069] 2) Construct a knowledge graph for process industry technologies based on the triples in the knowledge base; add priorities to the triples according to the adjustment order each time, and construct a priority-based process industry knowledge graph model. The obtained triples {FX_SX_BCYW, higher than normal, FX_SX_ZFKD3-2, 0}, {FX_SX_BCYW, higher than normal, FX_SX_ZFKD2-1, 1}, {FX_SX_BCYW, higher than normal, FX_CX_CQL3-1, 2}, and {FX_SX_BCYW, higher than normal, FX_CX_ZFKD3-1, 2} constitute the knowledge graph for process industry technologies.

[0070] 3) Based on real-time and historical data of collected equipment tags and production product indicators, the knowledge graph reasoning method, when optimizing process industries, requires adjusting the parameters of the tags. The tags and specific parameters to be adjusted are written into the corresponding equipment, and the adjustment of relevant process parameters is locked, allowing sufficient reaction time during production. Then, reasoning and optimization adjustments continue. Finally, the associated information is output for operators to check and provide feedback promptly.

[0071] The initial triggering sequence is shown in Table 1:

[0072]

[0073] After the first triggering of the triplet in Table 1, the priority of the corresponding triplet is modified, and the order of the next triggering is shown in Table 2.

[0074]

[0075] Table 2 Second Trigger Sequence

[0076] Upon the third trigger, two triples are obtained: {FX_SX_BCYW, higher than normal, FX_CX_LD} and {FX_SX_BCYW, higher than normal, FX_CX_ZFKD3-1}. For {FX_SX_BCYW, higher than normal, FX_CX_ZFKD3-1}, the value of FX_CX_ZFKD3 is decremented by 1 and written to device tuning. For {FX_SX_BCYW, higher than normal, FX_CX_LD}, since the tail entity is not a tag setting, FX_CX_LD is used as the head entity, and the search continues downwards.

[0077] 4) Using the existing set of triples, derive new triples to complete the knowledge graph. During the first graph completion, m is always 0; the calculated distances for the head entity FX_SX_BCYW and the relation "above normal" are shown in Table 3.

[0078] entity distance 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

[0079] Table 3 Calculation of triplet distance (first time)

[0080] At this point, the triplet (FX_SX_BCYW, higher than normal, FX_SX_BP) was inferred. However, experts determined through experience and process that this triplet was an incorrect triplet, with an error count of 1. Then, by continuing the inference using formula (2), a new calculated distance was obtained, as shown in Table 4:

[0081] entity distance 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

[0082] Table 4 Calculation of triplet distance (2nd time)

[0083] The number of incorrect inferences for the triple (FX_SX_BCYW, above normal, FX_SX_BP) was 2. Completion was then performed, and the new calculated distances are shown in Table 5.

[0084] entity distance 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

[0085] Table 5 Calculation of triplet distance (3rd time)

[0086] At this point, a new triplet (FX_SX_BCYW, higher than normal, FX_JX_CQL1+1) was obtained, which matches the correct fact.

[0087] It can be seen that the obtained triplet optimization strategy meets expectations, verifying that knowledge graph completion has a good effect.

Claims

1. A process industry process optimization method based on knowledge graph reasoning and completion, characterized in that, The method comprises the following steps: The equipment in the process design of the process industry, the equipment label, the production product index, and the equipment label adjustment amount are taken as entities to construct a triple structure of head entity-relation-tail entity; Based on the triple structure, a knowledge graph for the process of the process industry is constructed; Real-time data and historical data of the equipment label and the production product index are obtained, and based on the knowledge graph, the process parameters are adjusted to realize process optimization; The knowledge graph is completed based on the existing triple set; The knowledge graph for the process of the process industry is constructed based on the triple structure, specifically: According to the set optimization order of the trigger amount in the process of the process industry, a priority is added in the triple to make each triple (h, r, t) become a corresponding quadruple (h, r, t, o), wherein h represents the head entity, r represents the trigger relation between the head entity and the tail entity, t represents the tail entity, and o represents the priority of h and r pointing to t; a plurality of quadruples constitute the knowledge graph for the process of the process industry; The adjustment of the process parameters comprises the following steps: 1) Real-time data and historical data of the equipment label and the production product index are obtained, and the data is put into the quadruple (h, r, t, o); 2) The trigger relation r' of the last cycle is obtained, if r'≠r, then the priority of all quadruples diffused by the last cycle route is reset, and step 3) is executed; otherwise, step 3) is executed according to the current priority, wherein the route is a route composed of triples executed in the knowledge graph, that is, the tail entity of the previous triple is the head entity of the next triple; 3) In the quadruple, the same head entity h and trigger relation r correspond to a plurality of different tail entities t, 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 step 3) is the equipment label adjustment amount, the equipment parameters are optimized, that is, step 5) is executed; otherwise, the tail entity t1 is taken as the head entity in the quadruple, and step 2) is returned; if the tail entity t1 is the same as the current head entity h, the execution of the route is terminated; 5) The label corresponding to the tail entity t1 is obtained, the current value v1 of the label is obtained from the real-time data, and the value v2 to be adjusted is obtained from the equipment label adjustment amount corresponding to the label; v2 is compared with the upper and lower thresholds of the label, if the value adjusted by v2 is within the threshold, v2 is written to the equipment; otherwise, step 2) is returned; 6) The priority of the adjusted entity is reduced, and the reaction time is waited, if there is a new trigger knowledge graph event, step 1) is returned, otherwise, step 7) is executed; 7) If the adjusted tail entity label still has a quadruple with it as the head entity, when the equipment monitoring value and the process production process value continue for a period of time, the tail entity t is adjusted back to the original value.

2. The process industry process optimization method based on knowledge graph reasoning and completion according to claim 1, characterized in that, The relation is a relation set extracted from the process adjustment history and the abnormal treatment history, and at least one of the following relations is included: the relation between the equipment and the equipment label, the relation between the equipment label and the equipment label adjustment amount, the relation between the production product index and the equipment label adjustment amount, and the relation between the equipment label and the production product index.

Images