A knowledge graph-based auxiliary decision-making method for dam safety

Abstract

The present invention relates to a dam safety decision-making assistance method based on a knowledge graph. The method comprises: obtaining a description of the hidden danger problems occurring in the current dam; based on the description of the hidden danger problems of the current dam, extracting N hidden danger problem entities with the highest semantic similarity from the hidden danger problem set as candidate hidden danger problem entities, N≥2; based on the candidate hidden danger problem entities, combined with the dam hidden danger problem knowledge graph, forming dam hidden danger problem cases corresponding to each candidate hidden danger problem entity; based on the dam hidden danger problem cases corresponding to each candidate hidden danger problem entity, combined with a preset case evaluation system, obtaining the recommendation coefficients of each dam hidden danger problem case, and determining the recommended case for decision-making assistance based on the recommendation coefficient. The present invention can quickly match similar dam hidden danger problem cases when hidden danger problems occur in the dam through semantic matching of dam hidden danger problems and knowledge graph of dam hidden danger problems, and present them in a combination of sub-graph and text.

Description

Technical Field

[0001] The present invention relates to a knowledge graph-based dam safety decision-making auxiliary method, which is applicable to the field of dam safety management. Background Art

[0002] Dam damage and collapse is a gradual process from quantitative change to qualitative change. Discovering potential problems of dams in advance and repairing them is the most effective way to prevent dam collapse accidents. In the practice of dam safety management, experts with rich engineering experience and professional knowledge can often quickly give corresponding engineering disposal measures for hidden dangers. However, some front-line dam safety managers do not have this ability due to lack of engineering experience. It is necessary to effectively organize and utilize historical dam hidden danger cases to assist dam managers in making accurate decisions. Summary of the invention

[0003] The technical problem to be solved by the present invention is: to provide a dam safety decision-making assistance method based on knowledge graph in response to the above-mentioned problems.

[0004] The technical solution adopted by the present invention is: a dam safety decision-making auxiliary method based on knowledge graph, characterized by comprising:

[0005] Obtain descriptions of potential hazards currently occurring at the dam;

[0006] Based on the description of the hidden dangers of the current dam, N hidden danger problem entities with the highest semantic similarity are extracted from the hidden danger problem set as candidate hidden danger problem entities, N ≥ 2;

[0007] Based on the candidate hidden danger problem entities and combined with the dam hidden danger problem knowledge graph, a dam hidden danger problem case corresponding to each candidate hidden danger problem entity is formed;

[0008] Based on the dam hidden danger problem cases corresponding to each candidate hidden danger problem entity and combined with the preset case evaluation system, the recommendation coefficients of each dam hidden danger problem case are obtained, and the recommended cases for auxiliary decision-making are determined based on the recommendation coefficients;

[0009] The hidden danger problem set is constructed based on the knowledge graph of dam hidden danger problems; the evaluation indicators of the case evaluation system include one or more of semantic similarity, dam type similarity, engineering level similarity, treatment effect, and plan detail.

[0010] Based on the description of the hidden danger problem of the current dam, N hidden danger problem entities with the highest semantic similarity are extracted from the hidden danger problem set as candidate hidden danger problem entities, including:

[0011] Input the description of the hidden danger problem of the current dam and each hidden danger problem entity in the hidden danger problem set into the trained dam safety hidden danger semantic matching model to obtain the semantic similarity between the description of the hidden danger problem of the current dam and each hidden danger problem entity;

[0012] Based on the semantic similarity between the current dam hidden danger problem description and each hidden danger problem entity, N hidden danger problem entities with the highest semantic similarity are extracted from the hidden danger problem set as candidate hidden danger problem entities.

[0013] The dam safety hazard semantic matching model is constructed based on the twin BERT network.

[0014] Based on the candidate hidden danger problem entities, combined with the dam hidden danger problem knowledge graph, a dam hidden danger problem case corresponding to each candidate hidden danger problem entity is formed, including:

[0015] For candidate hidden danger problem entities, retrieve the entity labels related to them in the dam hidden danger problem knowledge graph to form a dam hidden danger problem case sub-graph;

[0016] For the entities in the dam hidden danger problem case sub-graph, the case-related text content is retrieved by querying the entity attributes in the knowledge graph to form a complete text representation of the dam hidden danger problem case.

[0017] The dam hidden danger problem cases corresponding to each candidate hidden danger problem entity are combined with a preset case evaluation system to obtain recommendation coefficients for each dam hidden danger problem case, and the recommended cases for auxiliary decision-making are determined based on the recommendation coefficients, including:

[0018] ;

[0019] ;

[0020] ;

[0021] ;

[0022] ;

[0023] in, f It is the recommended coefficient for the case of dam hidden danger problem; s The semantic similarity coefficient between the candidate hidden danger problem entity corresponding to the dam hidden danger problem case and the current dam hidden danger problem description; t is the dam type similarity coefficient between the current dam and the case dam; g is the engineering level similarity coefficient between the current dam and the case dam; For the current dam project level; is the grade of the dam project in the case; e is the score of the hidden danger treatment effect of the case; m is the score of the detailed level of the plan in the case; The number of words in the case text.

[0024] The entity tags in the dam hidden danger problem knowledge graph include hidden danger problems, abnormal performance, hidden danger causes, disposal measures, disposal effects, knowledge summary, dam, and dam type.

[0025] A dam safety decision support system based on knowledge graph, characterized by comprising:

[0026] The problem acquisition module is used to obtain the description of the hidden dangers of the current dam;

[0027] The semantic matching module is used to extract N hidden danger problem entities with the highest semantic similarity from the hidden danger problem set based on the description of the hidden danger problem of the current dam as candidate hidden danger problem entities, N ≥ 2;

[0028] A case formation module is used to form dam hidden danger problem cases corresponding to each candidate hidden danger problem entity based on the candidate hidden danger problem entity and in combination with the dam hidden danger problem knowledge graph;

[0029] The case recommendation module is used to obtain the recommendation coefficients of the dam hidden danger problem cases based on the dam hidden danger problem cases corresponding to each candidate hidden danger problem entity and in combination with the preset case evaluation system, and determine the recommended cases for auxiliary decision-making based on the recommendation coefficients;

[0030] The hidden danger problem set is constructed based on the knowledge graph of dam hidden danger problems; the evaluation indicators of the case evaluation system include one or more of semantic similarity, dam type similarity, engineering level similarity, treatment effect, and plan detail.

[0031] A computer storage medium stores a computer program that can be executed by a processor, characterized in that when the computer program is executed, the steps of the knowledge graph-based dam safety decision-making support method are implemented.

[0032] A dam safety decision-making support device comprises a memory and a processor, wherein the memory stores a computer program that can be executed by the processor, and wherein the computer program implements the steps of the knowledge graph-based dam safety decision-making support method when executed.

[0033] The beneficial effects of the present invention are as follows: through semantic matching of dam hidden danger problems and knowledge graphs of dam hidden danger problems, the present invention can quickly match similar dam hidden danger problem cases when hidden danger problems occur in the dam, and present them in a combination of sub-graphs and texts. The present invention determines the recommendation coefficient of the case from aspects such as semantic similarity, dam type, engineering level, treatment effect, and detail of the plan, and sorts similar dam hidden danger problem cases to construct an accurate recommendation model for dam hidden danger problem cases, providing decision support for dam safety managers.

[0034] The present invention is highly consistent with expert reasoning logic, stores historical dam hidden danger problem cases and experiences in the form of knowledge graphs, and uses the BERT model and knowledge reasoning method to perform semantic analysis and matching of dam hidden danger problems, providing a new idea for accurate decision-making support for dam safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A flowchart of an embodiment.

[0036] Figure 2 It is a block diagram of the semantic matching model of dam hidden danger problem in the embodiment.

[0037] Figure 3 This is a sub-map of the dam hidden danger problem case in the embodiment. DETAILED DESCRIPTION

[0038] Example 1: Figure 1 As shown, this embodiment is a dam safety decision-making support method based on a knowledge graph, which pre-constructs a dam hidden danger problem knowledge graph and a hidden danger problem set, and pre-trains a dam hidden danger problem semantic matching model, etc.

[0039] In this example, a knowledge system of dam hidden danger problems is formulated, cases of dam hidden danger problems are collected, and a knowledge graph of dam hidden danger problems is constructed. The specific steps are as follows:

[0040] (1) Based on the business needs of dam safety decision-making assistance, a knowledge system of dam hidden dangers is formulated, in which entity tags include but are not limited to hidden dangers, abnormal manifestations, hidden danger causes, disposal measures, disposal effects, knowledge summary, dam, and dam type.

[0041] In this embodiment, the attribute content of the hidden danger problem entity includes the name of the hidden danger problem and a detailed description of the hidden danger problem; the attributes of the abnormal performance, hidden danger cause, and disposal measures entities include a detailed description of the corresponding content; the attribute content of the disposal effect entity includes a description of the effect after the hidden danger problem is disposed and the disposal effect level. The disposal effect level is used to grade the disposal effect, which is divided into four levels: significant effect, good effect, general effect, and poor effect; the attribute content of the knowledge summary entity includes the lessons learned from summarizing and generalizing the case; the attribute of the dam entity is the name of the dam where the hidden danger problem occurs; the attribute of the dam type entity is the dam type of the dam where the hidden danger problem occurs, and the categories include concrete gravity dam, concrete arch dam, homogeneous earth dam, concrete panel rockfill dam, etc.

[0042] (2) According to the knowledge system of dam hidden danger problems, a large number of dam hidden danger problem cases were extracted from professional reports, professional books, standards and specifications, relevant websites and other materials.

[0043] In this embodiment, professional reports include but are not limited to dam safety periodic inspection reports, dam safety level registration reports, dam safety daily inspection reports, dam safety special inspection reports, etc.; the professional books include paper books and electronic books; the case extraction methods include manual extraction and intelligent extraction, among which the intelligent extraction methods include but are not limited to OCR recognition, UIE extraction model, etc.

[0044] (3) Referring to the entity labels in the knowledge system of dam hidden danger problems, for each dam hidden danger problem case, the text is split and named in the form of hidden danger problem, abnormal manifestation, hidden danger cause, disposal measures, disposal effect, knowledge summary, dam, and dam type to form candidate entities of the knowledge graph. The candidate entities are then associated and corresponded in the form of (hidden danger problem → abnormal manifestation), (hidden danger problem → abnormal cause), (hidden danger problem → disposal measures), (hidden danger problem → knowledge summary), (disposal measures → disposal effect), (dam → hidden danger problem), and (dam → dam type) to form a large number of dam hidden danger problem triplets.

[0045] (4) With the help of the Py2neo toolkit, all dam hidden danger problem triplets are imported into the Neo4j database in batches using Python programming to complete the construction of the dam hidden danger problem knowledge graph.

[0046] In this embodiment, all hidden danger problem entities are retrieved from the knowledge graph of dam hidden danger problems, and the entity names are constructed into a hidden danger problem set as key information for semantic matching. The Cypher search command and the hidden danger problem set are as follows:

[0047] (1)

[0048] (2)

[0049] in, Represents a set of potential problems. Represents the number of hidden danger problem entities in the dam hidden danger problem knowledge graph.

[0050] In this embodiment, secondary training is performed on the dam hidden danger problem sentence based on the twin BERT network to form a semantic matching model for the dam hidden danger problem. The specific steps are as follows:

[0051] (a) With { A , B , s} format to artificially construct a large amount of training data, A , B They represent different statements about dam hidden dangers. s Representation Statement A and statements B The semantic similarity coefficient of s The closer it is to 1, the more likely the statement A and statements B The higher the semantic similarity.

[0052] (b) Investigate various open source BERT architecture models, and select BERT pre-trained models with better understanding of Chinese semantics as the base model for secondary training of dam hidden danger problem sentences. The BERT pre-trained model is a BERT model or BERT variant model trained based on various data, including but not limited to the bge-large-zh-v1.5 model, bert-base-chinese model, simbert-base-chinese model, etc.

[0053] (c) The training data is divided into training samples, verification samples, and test samples in a ratio of 8:1:1, and the BERT base model is trained again to form a dedicated BERT model suitable for semantic analysis of dam hidden danger problems.

[0054] (d) Construct a twin BERT network structure including input layer, feature extraction layer, and output layer to form a special semantic matching model for dam hidden danger problems, such as Figure 2 As shown. The input layer is input to the sentence A and statements B The feature extraction layer includes a BERT model and a pooling strategy. The BERT model uses the above-mentioned dedicated BERT model obtained through secondary training. The pooling strategy can select average pooling or maximum pooling. The output layer is the output sentence. A and statements BThe semantic similarity coefficient.

[0055] The dam safety decision-making assistance method based on the knowledge graph in this embodiment specifically includes the following steps:

[0056] S1. Obtain description of the hidden dangers of the current dam.

[0057] S2. Based on the description of the hidden dangers of the current dam, the 10 hidden danger entities with the highest semantic similarity are extracted from the hidden danger set.

[0058] Describe the hidden dangers of the current dam and the hidden danger entities c in the hidden danger set C i , input the trained dam safety hazard semantic matching model to obtain the semantic similarity coefficient s between the current dam hazard problem description and each hazard problem entity; and extract the top 10 items with the highest similarity coefficient as candidate hazard problem entities.

[0059] S3. Based on the candidate hidden danger problem entities and combined with the dam hidden danger problem knowledge graph, a dam hidden danger problem case corresponding to each candidate hidden danger problem entity is formed.

[0060] For each candidate hidden danger problem entity, the abnormal performance, hidden danger cause, knowledge summary, disposal measures, disposal effect, dam, and dam type entities related to it are retrieved through single-hop and multi-hop query in the dam hidden danger problem knowledge graph to form a dam hidden danger problem case sub-graph. For example, Figure 3 shown.

[0061] For the entities in the dam hidden danger problem case sub-graph, the case-related text content is retrieved by querying the entity attributes in the knowledge graph to form a complete text representation of the dam hidden danger problem case.

[0062] S4. Based on the dam hidden danger problem cases corresponding to each candidate hidden danger problem entity and combined with the preset case evaluation system, the recommendation coefficients of each dam hidden danger problem case are obtained, and the recommended cases for auxiliary decision-making are determined based on the recommendation coefficients.

[0063] For the similar dam hidden danger problem cases matched in step S3, they are ranked in terms of semantic similarity, dam type similarity, engineering level similarity, treatment effect, and solution detail, and a precise recommendation model for dam hidden danger problem cases is constructed. The specific steps are as follows:

[0064] ①, semantic similarity is the semantic similarity coefficient between the hidden danger entity corresponding to the dam hidden danger problem case matched in step S3 and the current dam hidden danger problem description s , the value range is [0, 1], the semantic similarity coefficient s The larger the value, the higher the semantic similarity between the two sentences.

[0065] ② Dam type and engineering level are used to judge the matching degree between the current dam and the dam with similar hidden danger cases, using dam type coefficient and engineering grade coefficient If the current dam has the same dam type as the case dam, then the dam type coefficient ; If the dam type is different, then , as shown below:

[0066] (3)

[0067] Dam engineering is classified into five levels, from level 1 to level 5, which can be expressed parameterized as , for the current dam project level Dam projects with similar hidden dangers , the engineering grade coefficients of the two are defined as:

[0068] (4)

[0069] That is, if the current dam has the same engineering grade as the dam with similar hidden dangers, the engineering grade coefficient If the engineering level of the current dam is different from that of the dam with similar hidden dangers, the engineering level coefficient For every one-grade deviation in engineering grade, the engineering grade coefficient will be reduced by 0.1.

[0070] ③. The hidden danger treatment effect and the level of detail of the plan are used to judge the availability of candidate similar hidden danger problem cases, using the hidden danger treatment effect parameters and the scenario detail level parameter As described in step S1, the hidden danger treatment effect is divided into four levels: significant effect, good effect, general effect and poor effect. The corresponding hidden danger treatment effect parameters are set. They are 1, 0.8, 0.6, and 0.4 respectively, as shown below:

[0071] (5)

[0072] Scenario detail parameters Word count of similar hidden danger case text The more words a text has, the more detailed the case plan is and the greater its reference value is. Divide into four intervals: [1000, ∞), [500,1000), [200, 500), [0, 200), and set the corresponding scheme detail parameters They are 1, 0.85, 0.7, and 0.5 respectively, as shown below:

[0073] (6)

[0074] ④. Comprehensively considering the similarity of hidden dangers, dam type, project scale, treatment effect, and details of the plan, the recommendation coefficient of each similar dam hidden danger case matched is calculated as follows:

[0075] (7)

[0076] ⑤. For the dam hidden danger problem case matched in step S3, the case recommendation coefficient calculated in formula (7) is Sort from large to small to achieve accurate recommendation of the most similar dam hazard cases.

[0077] Embodiment 2: This embodiment is a dam safety decision support system based on knowledge graph, including: a question acquisition module, a semantic matching module, a case formation module and a case recommendation module, etc.

[0078] In this example, the problem acquisition module is used to obtain the description of the hidden danger problems of the current dam; the semantic matching module is used to extract the N hidden danger problem entities with the highest semantic similarity from the hidden danger problem set based on the description of the hidden danger problems of the current dam as candidate hidden danger problem entities, N ≥ 2.

[0079] In this embodiment, the case formation module is used to form dam hidden danger problem cases corresponding to each candidate hidden danger problem entity based on the candidate hidden danger problem entity and in combination with the dam hidden danger problem knowledge graph; the case recommendation module is used to obtain the recommendation coefficients of each dam hidden danger problem case based on the dam hidden danger problem case corresponding to each candidate hidden danger problem entity and in combination with the preset case evaluation system, and determine the recommended case for auxiliary decision-making based on the recommendation coefficient;

[0080] In this embodiment, the hidden danger problem set is constructed based on the knowledge graph of dam hidden danger problems; the evaluation indicators of the case evaluation system include semantic similarity, dam type similarity, engineering level similarity, treatment effect, and plan detail.

[0081] Embodiment 3: This embodiment is a computer storage medium, on which a computer program that can be executed by a processor is stored. When the computer program is executed, the steps of the dam safety decision-making support method based on the knowledge graph in Embodiment 1 are implemented.

[0082] Example 4: This example is a dam safety decision-making support device, which has a memory and a processor. The memory stores a computer program that can be executed by the processor. When the computer program is executed, the steps of the dam safety decision-making support method based on the knowledge graph in Example 1 are implemented.

Claims

1. A dam safety decision-making assistance method based on knowledge graph, characterized in that: include: Obtain descriptions of potential hazards currently occurring at the dam; Based on the description of the hidden dangers of the current dam, N hidden danger problem entities with the highest semantic similarity are extracted from the hidden danger problem set as candidate hidden danger problem entities, N ≥ 2; Based on the candidate hidden danger problem entities and combined with the dam hidden danger problem knowledge graph, a dam hidden danger problem case corresponding to each candidate hidden danger problem entity is formed; Based on the dam hidden danger problem cases corresponding to each candidate hidden danger problem entity and combined with the preset case evaluation system, the recommendation coefficients of each dam hidden danger problem case are obtained, and the recommended cases for auxiliary decision-making are determined based on the recommendation coefficients; The hidden danger problem set is constructed based on the knowledge graph of dam hidden danger problems; the evaluation indicators of the case evaluation system include one or more of semantic similarity, dam type similarity, engineering level similarity, treatment effect, and solution detail; Based on the candidate hidden danger problem entities, combined with the dam hidden danger problem knowledge graph, a dam hidden danger problem case corresponding to each candidate hidden danger problem entity is formed, including: For candidate hidden danger problem entities, retrieve the entity labels related to them in the dam hidden danger problem knowledge graph to form a dam hidden danger problem case sub-graph; For entities in the dam hidden danger problem case sub-graph, the case-related text content is retrieved by querying entity attributes in the knowledge graph to form a complete text representation of the dam hidden danger problem case; The dam hidden danger problem cases corresponding to each candidate hidden danger problem entity are combined with a preset case evaluation system to obtain recommendation coefficients for each dam hidden danger problem case, and the recommended cases for auxiliary decision-making are determined based on the recommendation coefficients, including: f=s+t+g+e+m g=1-0.1×(d p -d i ) Among them, f is the recommendation coefficient of the dam hidden danger problem case; s is the semantic similarity coefficient between the candidate hidden danger problem entity corresponding to the dam hidden danger problem case and the current dam hidden danger problem description; t is the dam type similarity coefficient between the current dam and the case dam; g is the engineering level similarity coefficient between the current dam and the case dam; d p is the current dam project level; d i is the grade of the dam project in the case; e is the score of the hidden danger treatment effect of the case; m is the score of the detail level of the plan of the case; w is the number of words in the text of the case.

2. The dam safety decision-making assistance method based on knowledge graph according to claim 1 is characterized in that: Based on the description of the hidden danger problem of the current dam, N hidden danger problem entities with the highest semantic similarity are extracted from the hidden danger problem set as candidate hidden danger problem entities, including: Input the description of the hidden danger problem of the current dam and each hidden danger problem entity in the hidden danger problem set into the trained dam safety hidden danger semantic matching model to obtain the semantic similarity between the description of the hidden danger problem of the current dam and each hidden danger problem entity; Based on the semantic similarity between the current dam hidden danger problem description and each hidden danger problem entity, N hidden danger problem entities with the highest semantic similarity are extracted from the hidden danger problem set as candidate hidden danger problem entities.

3. The dam safety decision-making assistance method based on knowledge graph according to claim 2 is characterized in that: The dam safety hazard semantic matching model is constructed based on the twin BERT network.

4. The dam safety decision-making assistance method based on knowledge graph according to claim 1 is characterized in that: The entity tags in the dam hidden danger problem knowledge graph include hidden danger problems, abnormal performance, hidden danger causes, disposal measures, disposal effects, knowledge summary, dam, and dam type.

5. A knowledge graph-based dam safety decision support system, characterized in that: include: The problem acquisition module is used to obtain the description of the hidden dangers of the current dam; The semantic matching module is used to extract N hidden danger problem entities with the highest semantic similarity from the hidden danger problem set based on the description of the hidden danger problem of the current dam as candidate hidden danger problem entities, N ≥ 2; A case formation module is used to form dam hidden danger problem cases corresponding to each candidate hidden danger problem entity based on the candidate hidden danger problem entity and in combination with the dam hidden danger problem knowledge graph; The case recommendation module is used to obtain the recommendation coefficients of the dam hidden danger problem cases based on the dam hidden danger problem cases corresponding to each candidate hidden danger problem entity and in combination with the preset case evaluation system, and determine the recommended cases for auxiliary decision-making based on the recommendation coefficients; The hidden danger problem set is constructed based on the knowledge graph of dam hidden danger problems; the evaluation indicators of the case evaluation system include one or more of semantic similarity, dam type similarity, engineering level similarity, treatment effect, and solution detail; Based on the candidate hidden danger problem entities, combined with the dam hidden danger problem knowledge graph, a dam hidden danger problem case corresponding to each candidate hidden danger problem entity is formed, including: For candidate hidden danger problem entities, retrieve the entity labels related to them in the dam hidden danger problem knowledge graph to form a dam hidden danger problem case sub-graph; For entities in the dam hidden danger problem case sub-graph, the case-related text content is retrieved by querying entity attributes in the knowledge graph to form a complete text representation of the dam hidden danger problem case; The dam hidden danger problem cases corresponding to each candidate hidden danger problem entity are combined with a preset case evaluation system to obtain recommendation coefficients for each dam hidden danger problem case, and the recommended cases for auxiliary decision-making are determined based on the recommendation coefficients, including: f=s+t+g+e+m g=1-0.1×(d p -d i ) Among them, f is the recommendation coefficient of the dam hidden danger problem case; s is the semantic similarity coefficient between the candidate hidden danger problem entity corresponding to the dam hidden danger problem case and the current dam hidden danger problem description; t is the dam type similarity coefficient between the current dam and the case dam; g is the engineering level similarity coefficient between the current dam and the case dam; d p is the current dam project level; d i is the grade of the dam project in the case; e is the score of the hidden danger treatment effect of the case; m is the score of the detail level of the plan of the case; w is the number of words in the text of the case.

6. A computer storage medium having stored thereon a computer program executable by a processor, characterized in that: When the computer program is executed, the steps of the dam safety decision-making support method based on knowledge graph described in any one of claims 1 to 4 are implemented.

7. A dam safety decision-making support device, comprising a memory and a processor, wherein the memory stores a computer program executable by the processor, characterized in that: When the computer program is executed, the steps of the dam safety decision-making support method based on knowledge graph described in any one of claims 1 to 4 are implemented.

Images