Mine ventilation intelligent decision-making method based on knowledge representation learning

Abstract

The invention relates to the technical field of mine ventilation, in particular to a mine ventilation intelligent decision-making method based on knowledge representation learning, and mainly solves the technical problems that an existing mine ventilation intelligent decision-making method is difficult to make decisions autonomously and low in intelligent degree. The method comprises the following steps: S1, constructing a ventilation knowledge graph; s2, constructing a ventilation knowledge ontology; s3, constructing a TransE model, and training to obtain a target model; s4, performing vectorization representation of different dimensions on the ventilation knowledge by adopting a target model, and selecting an optimal embedding dimension; and S5, based on the optimal embedding dimension and the ventilation knowledge ontology, performing intelligent reasoning on the input ventilation knowledge through the target model, and generating a control instruction of the ventilation system according to a reasoning result. According to the method, the self-learning, self-adaption and autonomous decision-making capabilities of the mine ventilation system can be effectively enhanced, so that the method can be suitable for intelligent management scenes such as ventilation abnormity early warning, risk judgment and control strategy automatic generation.

Description

Technical Field

[0001] This invention relates to the field of mine ventilation technology, and in particular to an intelligent decision-making method for mine ventilation based on knowledge representation learning. Background Technology

[0002] Mine ventilation systems are a core component of safe coal mine production, and their intelligent decision-making is becoming a key area of ​​research in the industry.

[0003] Existing intelligent decision-making methods for mine ventilation have the following shortcomings: First, although artificial intelligence technology has promoted the development of intelligent ventilation systems, existing systems still mainly rely on expert experience and rules, lacking the ability to deeply mine massive ventilation data and make autonomous decisions; Second, although current research has explored ventilation decision-making methods based on artificial intelligence, there are obvious shortcomings in knowledge representation and reasoning, resulting in low intelligence and difficulty in effectively processing complex ventilation data and achieving structured integration of knowledge.

[0004] Therefore, there is an urgent need for an intelligent decision-making method for mine ventilation that can make autonomous decisions and has a high degree of intelligence. Summary of the Invention

[0005] To overcome the technical shortcomings of existing intelligent decision-making methods for mine ventilation, such as difficulty in autonomous decision-making and low level of intelligence, this invention provides an intelligent decision-making method for mine ventilation based on knowledge representation learning.

[0006] The intelligent decision-making method for mine ventilation based on knowledge representation learning provided by this invention includes the following steps:

[0007] S1. Collect ventilation knowledge and construct a ventilation knowledge graph;

[0008] S2. Apply semantic constraints to the ventilation knowledge graph to form a ventilation knowledge ontology;

[0009] S3. Construct the TransE model and train the TransE model using the ventilation knowledge graph as the dataset to obtain the target model;

[0010] S4. Use the target model to vectorize ventilation knowledge in different dimensions and select the optimal embedding dimension;

[0011] S5. Based on the optimal embedding dimension and ventilation knowledge ontology, intelligent reasoning is performed on the input ventilation knowledge through the target model, and control instructions for the ventilation system are generated according to the reasoning results.

[0012] Optionally, step S1 includes the following sub-steps:

[0013] S11. Extract ventilation knowledge from semi-structured and unstructured data;

[0014] S12. Represent the ventilation knowledge as triplet data;

[0015] S13. The ventilation knowledge graph is formed after fusion.

[0016] Optionally, the ventilation knowledge includes ventilation entities and ventilation relationships.

[0017] Optionally, the triplet data can be in the form of ventilation entity + ventilation relationship + ventilation entity.

[0018] Optionally, step S1 may further include sub-step S14: performing reasoning and decision-making based on the ventilation knowledge graph to complete data annotation.

[0019] Optionally, step S2 includes the following sub-steps:

[0020] S21. Define the type, semantic attributes, and constraint rules of the ventilation knowledge;

[0021] S22. Associate the ventilation knowledge with the sensor data to form the ventilation knowledge ontology.

[0022] Optionally, step S4 includes the following sub-steps:

[0023] S41. Conduct vector embedding experiments with different dimensions under the TransE model;

[0024] S42. Use evaluation metrics to compare performance across different dimensions;

[0025] S43. Select the optimal embedding dimension through performance comparison.

[0026] Optionally, the evaluation metrics are MRR, MR, and HITS@n.

[0027] The technical solution provided by this invention has the following advantages compared with the prior art:

[0028] This invention provides a mine ventilation intelligent decision-making method based on knowledge representation learning. Firstly, it integrates mechanisms such as graph construction, representation learning, and intelligent reasoning, breaking through the bottleneck of traditional systems relying on expert experience and rules. Secondly, through structured extraction and multi-source fusion of ventilation knowledge, it achieves unified modeling of procedural knowledge and sensor data, constructing a semantic knowledge network for ventilation safety scenarios. Thirdly, it completes vectorized learning based on the TransE model and embeds the optimal embedding dimension and ventilation knowledge ontology, improving the accuracy and generalization ability of knowledge representation. Fourthly, it utilizes distance and similarity metrics in the vector space for reasoning analysis, enabling automatic identification, judgment, and response command generation for ventilation events. Thus, the combination of these four aspects effectively enhances the self-learning, self-adaptation, and autonomous decision-making capabilities of mine ventilation systems, making it applicable to intelligent management scenarios such as ventilation anomaly early warning, risk assessment, and automatic generation of control strategies, demonstrating good versatility and promotional value. Attached Figure Description

[0029] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 A flowchart illustrating the decision-making method in an embodiment of the present invention;

[0032] Figure 2 This is a flowchart illustrating the construction process of the ventilation knowledge graph in an embodiment of the present invention.

[0033] Figure 3 A framework diagram illustrating the ventilation knowledge ontology in an embodiment of the present invention;

[0034] Figure 4 This diagram shows the entity representation result of the TransE model in this embodiment of the invention.

[0035] Figure 5 This diagram shows the entity representation result of the TransR model in this embodiment of the invention.

[0036] Figure 6 This diagram shows the entity representation result of the RESCAL model in this embodiment of the invention.

[0037] Figure 7 This diagram shows the entity representation result of the DistMult model in this embodiment of the invention.

[0038] Figure 8 This diagram shows the results of the ventilation entity similarity analysis in the embodiments of the present invention. Detailed Implementation

[0039] To better understand the above-mentioned objectives, features, and advantages of the present invention, the solutions of the present invention will be further described below. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0040] Many specific details are set forth in the following description in order to provide a full understanding of the invention, but the invention may also be practiced in other ways different from those described herein; obviously, the embodiments in the specification are only some embodiments of the invention, and not all embodiments.

[0041] The following is combined Figures 1 to 8 Specific embodiments of the present invention will be described in detail below.

[0042] This embodiment provides a mine ventilation intelligent decision-making method based on knowledge representation learning, such as... Figure 1 As shown, it includes steps S1 to S5.

[0043] S1. Collect ventilation knowledge and construct a ventilation knowledge graph.

[0044] like Figure 2 As shown, step S1 includes the following sub-steps:

[0045] S11. Extract ventilation knowledge from semi-structured and unstructured data;

[0046] Semi-structured data includes, but is not limited to, mine accident cases and coal mine safety websites;

[0047] Unstructured data includes, but is not limited to, coal mine safety regulations, mine ventilation documents, and mine ventilation-related policies;

[0048] Ventilation knowledge includes ventilation entities and ventilation relationships. Ventilation entities include, but are not limited to, ventilation events, ventilation facilities, ventilation data, and ventilation relationships, including but not limited to, monitoring, judgment, and decision-making.

[0049] S12. Represent ventilation knowledge as triplet data;

[0050] This means that the extracted ventilation knowledge is transformed into a standardized triplet structure, which is a grammatical-level structuring.

[0051] The form of ternary data is ventilation entity + ventilation relationship + ventilation entity, such as wind speed exceeding limit + belonging to + ventilation event, or wind speed exceeding limit + located at + mining face;

[0052] S13. After fusion, a ventilation knowledge graph is formed;

[0053] Fusion is the semantic merging of ventilation knowledge from different sources to resolve knowledge redundancy and conflict, making it a consistent knowledge network, which belongs to semantic level integration; fusion and representation are logically continuous but different levels of processing stages.

[0054] Step S14: Based on the ventilation knowledge graph, reason and make decisions about ventilation knowledge to complete data annotation;

[0055] The reasoning is based on the constructed ventilation knowledge graph to perform rule reasoning and semantic verification, which is used to verify the rationality of knowledge extraction and fusion. The knowledge relationships output by the reasoning will help generate and verify the labeled dataset.

[0056] S2. Apply semantic constraints to the ventilation knowledge graph to form a ventilation knowledge ontology.

[0057] Step S2 includes the following sub-steps:

[0058] S21. Define the types, semantic attributes, and constraint rules of ventilation knowledge;

[0059] S22. Link ventilation knowledge with sensor data to form a ventilation knowledge ontology;

[0060] The ventilation knowledge ontology is a high-level semantic structure built upon extracted ventilation entities and relationships. It defines and organizes concepts and constraints in the ventilation domain, including multiple core concept layers such as ventilation events, ventilation facilities, ventilation actions, ventilation locations, ventilation data, and rules. Figure 3 As shown.

[0061] The comparison between step S2 and step S1 is shown in the table below:

[0062]

[0063] S3. Construct the TransE model and train it using the ventilation knowledge graph as the dataset to obtain the target model.

[0064] By replacing the head ventilation entity or tail ventilation entity in the triplet data, negative sampling samples are generated, and a scoring function is established to evaluate positive and negative samples.

[0065] Define a loss function to evaluate the model's representation learning performance by comparing the scores of positive and negative samples;

[0066] The loss function used is a mean squared error loss function:

[0067] ;

[0068] in, , As a positive sample, For negative samples, For interval parameters;

[0069] The model was evaluated using MRR, MR, and HITS@n as evaluation metrics.

[0070] MRR, or Mean Reciprocal Rank, is expressed as follows:

[0071] ;

[0072] MR, or Mean Rank, is used to calculate the average rank of the prediction results.

[0073] Hits@n is used to count the proportion of correct answers that rank in the top n.

[0074] like Figures 4 to 7 The comparison shows that the TransE model's categories (such as ventilation events, actions, and locations) are clearly clustered in the vector space with well-defined boundaries, demonstrating good semantic separation capabilities. It performs best in terms of semantic discriminability, clustering features, and representation efficiency. While the TransR model can distinguish categories well, the distribution is relatively scattered. The RESCAL and DistMult models are too concentrated, and the boundaries between categories are blurred.

[0075] S4. Use the target model to vectorize ventilation knowledge in different dimensions and select the optimal embedding dimension.

[0076] Step S4 includes the following sub-steps:

[0077] S41. Conduct vector embedding experiments with different dimensions under the TransE model;

[0078] In this embodiment, the dimensions are set to 10, 20, 40, and 80.

[0079] S42. Use evaluation metrics to compare performance across different dimensions;

[0080] In this embodiment, MRR and HITS@n are selected as evaluation indicators, and n=10;

[0081] S43. Select the optimal embedding dimension through performance comparison;

[0082] The experimental results of this embodiment are shown in the table below:

[0083]

[0084] Experimental data shows that a dimension of 40 achieves the best balance between accuracy and computational cost, so this embodiment selects 40 as the optimal embedding dimension.

[0085] S5. Based on the optimal embedding dimension and ventilation knowledge ontology, intelligent reasoning is performed on the input ventilation knowledge through the target model, and control commands for the ventilation system are generated based on the reasoning results.

[0086] For example, when the sensor detects that the wind speed exceeds the limit, the system automatically determines that it is a ventilation failure, generates a control command to stop working, and notifies relevant personnel to handle the situation.

[0087] like Figure 8 As shown, the horizontal and vertical axes represent ventilation entities in the ventilation knowledge graph. The color intensity indicates the cosine similarity between entity vectors, and the darker the color, the more semantically similar they are.

[0088] from Figure 8 We can conclude the following: First, the darkest color near the diagonal indicates good model convergence, meaning that the self-similarity of each entity in the vector space is the highest, and the model can stably learn semantic representations. Second, some entities exhibit high similarity clusters, reflecting semantic correlations. For example, ventilation events such as "exceeding wind speed limits," "insufficient airflow," and "abnormal wind pressure" form a cluster, indicating that the model has captured their semantic approximation in safety procedures. Ventilation actions such as "stop working" and "cut off power" are spatially close to the above events, indicating that the model can identify the implicit association between events and actions. Third, the low similarity between entities of different categories and the clear distinction between categories indicate that the model can not only discover associations but also maintain the boundaries between entity categories.

[0089] Figure 8 The results show that knowledge representation learning based on the TransE model can effectively capture the semantic relevance between entities in the ventilation knowledge graph, realize the clustering of similar entities and the differentiation of dissimilar entities, and provide an accurate semantic foundation for subsequent intelligent reasoning and decision-making.

[0090] The above are merely specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Although detailed descriptions have been provided with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments, and they should all be covered within the protection scope of the claims.

Claims

1. A mine ventilation intelligent decision-making method based on knowledge representation learning, characterized in that, The method comprises the following steps: S1. Collect ventilation knowledge and build a ventilation knowledge graph; S2. Perform semantic constraint on the ventilation knowledge graph to form a ventilation knowledge ontology; S3. Build a TransE model, train the TransE model by taking the ventilation knowledge graph as a data set, and obtain a target model; S4. Use the target model to perform vectorization representation of ventilation knowledge in different dimensions, and select an optimal embedding dimension; S5. Based on the optimal embedding dimension and the ventilation knowledge ontology, perform intelligent reasoning on input ventilation knowledge through the target model, and generate a control instruction of a ventilation system according to a reasoning result.

2. The knowledge representation learning based mine ventilation intelligent decision method according to claim 1, characterized in that, Step S1 comprises the following sub-steps: S11. Extract ventilation knowledge from semi-structured data and unstructured data; S12. Represent the ventilation knowledge as triple data; S13. Form the ventilation knowledge graph after fusion.

3. The knowledge representation learning based mine ventilation intelligent decision method according to claim 2, characterized in that, The ventilation knowledge comprises ventilation entities and ventilation relationships.

4. The knowledge representation learning based mine ventilation intelligent decision method according to claim 2, characterized in that, The triple data is in the form of ventilation entity + ventilation relationship + ventilation entity.

5. The knowledge representation learning based mine ventilation intelligent decision method according to any one of claims 2 to 4, characterized in that, Step S1 further comprises a sub-step S14: reasoning and decision-making of ventilation knowledge based on the ventilation knowledge graph to complete data labeling.

6. The knowledge representation learning based mine ventilation intelligent decision method according to claim 1, characterized in that, Step S2 comprises the following sub-steps: S21. Define types, semantic attributes and constraint rules of the ventilation knowledge; S22. Associate the ventilation knowledge with sensor data to form the ventilation knowledge ontology.

7. The knowledge representation learning based mine ventilation intelligent decision method according to claim 1, characterized in that, Step S4 comprises the following sub-steps: S41. Set different dimensions for vector embedding experiments under the TransE model; S42. Compare performances in different dimensions by using evaluation indexes; S43. Select an optimal embedding dimension through performance comparison. 8.The method of claim 7, wherein, The evaluation indexes are MRR and HITS@n.

Images