CN117035424A - Mine safety situation collaborative sensing system and method based on anti-deduction learning - Google Patents
Mine safety situation collaborative sensing system and method based on anti-deduction learning Download PDFInfo
- Publication number
- CN117035424A CN117035424A CN202311064467.8A CN202311064467A CN117035424A CN 117035424 A CN117035424 A CN 117035424A CN 202311064467 A CN202311064467 A CN 202311064467A CN 117035424 A CN117035424 A CN 117035424A
- Authority
- CN
- China
- Prior art keywords
- data
- module
- mine safety
- safety situation
- machine learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000010801 machine learning Methods 0.000 claims abstract description 87
- 238000012549 training Methods 0.000 claims abstract description 38
- 238000012545 processing Methods 0.000 claims abstract description 32
- 230000004927 fusion Effects 0.000 claims abstract description 25
- 238000013528 artificial neural network Methods 0.000 claims description 33
- 238000011156 evaluation Methods 0.000 claims description 27
- 230000008447 perception Effects 0.000 claims description 24
- 239000003245 coal Substances 0.000 claims description 22
- 238000004519 manufacturing process Methods 0.000 claims description 22
- 238000004458 analytical method Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 7
- 230000001351 cycling effect Effects 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 3
- 238000007689 inspection Methods 0.000 claims description 3
- 238000013210 evaluation model Methods 0.000 abstract description 3
- 238000002372 labelling Methods 0.000 abstract 1
- 230000006870 function Effects 0.000 description 7
- 238000007500 overflow downdraw method Methods 0.000 description 6
- 238000013507 mapping Methods 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
- G06N5/022—Knowledge engineering; Knowledge acquisition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
- G06N5/041—Abduction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- General Engineering & Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Development Economics (AREA)
- Operations Research (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Quality & Reliability (AREA)
- Agronomy & Crop Science (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Primary Health Care (AREA)
- Molecular Biology (AREA)
- Mining & Mineral Resources (AREA)
- Marine Sciences & Fisheries (AREA)
- Animal Husbandry (AREA)
- Educational Administration (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The mine safety situation collaborative awareness system based on anti-deduction learning comprises a data acquisition module, a data processing module, a machine learning module, a logic reasoning module connection and a machine learning and logic reasoning fusion module; the method comprises the following steps: constructing a data set; dividing security situation level; after labeling part of the data set, training the model to form an initial classifier C; taking the rest unmarked data and inputting the rest unmarked data into C to obtain a pseudo tag of each data; comparing the pseudo tag of each data with the result output by the logic reasoning module, minimizing inconsistency, obtaining an anti-deduction tag through anti-deduction learning, and forming a new data set S; s is used for training the classifier C, model parameters are updated to form a new classifier C, S is input into the classifier C for prediction, and the step is repeated until the model converges or the label is not updated any more; and constructing a final mine safety situation grade evaluation model. The system and the method can realize accurate assessment of the security situation level.
Description
Technical Field
The invention belongs to the technical field of artificial intelligence, and particularly relates to a mine safety situation collaborative awareness system and method based on deduction learning.
Background
Coal is the main strategic prop energy of our country at present, and takes the dominant role in the energy structure all the time, and future coal resources can still be ballast and bottom guarantee of our energy system. Mine safety has been a very serious problem worldwide, especially in some developing countries, because of the lack of effective management and supervision mechanisms. The method is particularly important for evaluating the mine safety situation level and early warning in order to practically ensure the safety production work of the underground coal mine. Currently, artificial intelligence technology has been widely used in the field of mine safety. Machine learning can play a great role in mine safety, such as coal mine gas explosion prediction, coal mine worker behavior monitoring, coal mine transportation equipment prediction maintenance, mine safety management and the like. Machine learning can help predict and prevent potential safety hazards by analyzing and processing mine data, so that occurrence of mine accidents is reduced. However, due to the complexity and security of the mine environment, there is little currently large public data set concerning the mine. The data collected downhole has greater noise and the lower quality of the data set results in less accurate prediction of the machine learning model.
Disclosure of Invention
Aiming at the problems existing in the prior art, the invention provides a mine safety situation collaborative sensing system and a method based on deduction learning, wherein the system has a simple composition structure and low input cost, and can be used for accurately evaluating the underground safety situation level; according to the method, under the mine safety situation prediction methods driven by data and knowledge, mine safety situation prediction mechanisms based on deduction learning are researched, logic reasoning and machine learning characteristics can be comprehensively considered, the credibility of the two characteristics is modeled, a mine safety situation assessment mechanism combining the two methods can be constructed, the accuracy of safety situation level prediction can be remarkably improved, accordingly occurrence of mine accident can be effectively reduced, and the robustness of the system is enhanced.
In order to achieve the above purpose, the invention provides a mine safety situation collaborative awareness system based on deduction learning, which comprises a data acquisition module, a data processing module, a machine learning module, a logic reasoning module and a machine learning and logic reasoning fusion module:
the data acquisition module is connected with the data processing module and is used for constructing a perception data set for mine safety situation grade assessment according to perception data acquired by the environment and the sensors in the inspection robot in the mine safety production process and sending the perception data set to the data processing module;
the data processing module is respectively connected with the machine learning module and the logic reasoning module, is used for determining characteristic parameters for evaluating the mine safety situation level by combining expert experience and priori knowledge with a knowledge base, converts an original perception data set into a group of core characteristic data sets with obvious statistical significance by automatically constructing new characteristics, and respectively sends the core characteristic data sets to the machine learning module and the logic reasoning module;
the machine learning module is used for inputting the processed perception data into the sub-neural networks of different modes, integrating and fusing the prediction results of the sub-neural networks of each mode to obtain a final overall decision, inputting the overall decision into the mine safety situation level assessment pre-training model, finally constructing a data-driven safety situation perception model, generating a data-driven mine safety situation level assessment result according to training characteristics, and finally transmitting the mine safety situation level assessment result to the machine learning and logic reasoning fusion module;
the logic reasoning module is used for determining the basis for evaluating the mine safety situation level according to expert experience and priori knowledge, carrying out abstract expression in a first-order predicate logic form to form an expert knowledge base, constructing a knowledge-driven safety situation logic reasoning model based on analysis and reasoning of the knowledge base and historical data by a fuzzy logic algorithm, generating a knowledge-driven mine safety situation level evaluation result according to training characteristics, and sending the mine safety situation level evaluation result to the machine learning and logic reasoning fusion module;
the machine learning and logic reasoning fusion module is respectively connected with the machine learning module and the logic reasoning module, is used for comprehensively considering logic reasoning and machine learning characteristics, modeling the credibility of the machine learning module and the logic reasoning fusion module, generating a pseudo tag based on a prediction result of a neural network in the machine learning module, correcting the pseudo tag by utilizing the result of the logic reasoning module, updating a mine safety situation level evaluation pre-training model, and finally constructing and forming a mine safety situation collaborative sensing system based on anti-deduction learning.
Preferably, the mine safety situation level assessment pre-training model adopts a variation automatic encoder. Because the environment under the mine is limited, the collected data has larger noise, the dimension reduction processing is carried out on the data set containing the noise collected in the mine, the core characteristics for determining the data difference are extracted from the original high-dimension data, the noise can be effectively reduced and removed, and the generalization capability of the model is obviously improved.
In the invention, because the logic reasoning module is based on the fuzzy logic algorithm, different from the traditional Boolean logic algorithm, the rule used by the fuzzy logic algorithm is more flexible, and the ambiguity concept is mapped to the real number interval to calculate, so that the uncertainty of different degrees can be processed. The method also has certain processing capacity for data containing certain noise or inaccuracy, and can improve the robustness of the algorithm. Meanwhile, the fuzzy logic algorithm can handle a large number of uncertain or fuzzy situations and can be adjusted according to the needs, so that the sensing system is very flexible. In addition, the logic reasoning module is arranged and the machine learning module is also arranged, so that the advantages of the logic reasoning module and the machine learning module can be fully combined, the advantages of asynchronous data can be processed by utilizing a multi-mode late fusion method, and meanwhile, the logic reasoning module directly uses the original data to make up for the defect that the original data neglects low-level characteristics among a plurality of modes. Through the arrangement of the machine learning and logic reasoning fusion module, an effectively fused knowledge reasoning and data-driven mine safety situation collaborative sensing system can be formed, so that the characteristics of logic reasoning and machine learning can be fully and comprehensively considered, the evaluation result of the machine learning module can be detected and verified by utilizing the logic reasoning module, the mine safety situation level evaluation pre-training model can be updated, and further the mine safety situation collaborative sensing system based on anti-deduction learning can be formed, the safety level of a coal mine can be rapidly and accurately evaluated, and the safety production operation of the coal mine can be guaranteed. Because of the complex and varying mine environments, it may be difficult to use other machine learning methods to arrive at the correct conclusion with low reliability or low data volume. The system builds a knowledge-driven safety situation logic reasoning model through a fuzzy logic algorithm, builds a data-driven safety situation sensing model through machine learning algorithms such as a variation automatic encoder and the like, and based on anti-deduction learning, combines machine learning and logic reasoning prediction results, and updates a mine safety situation level assessment model after comprehensive consideration to obtain a final mine safety situation level assessment model, so that a correct conclusion can be obtained through deduction from a large amount of known information, and the system can be rapidly adapted to the uncertain and changed fields. The invention has simple structure, strong universality and easy implementation, effectively combines machine learning and logic reasoning together in a balanced and reciprocal mode, is beneficial to correcting perceived errors and enhances the robustness of the system. Aiming at the complex mine environment, the system can realize accurate assessment of the security situation level and can improve the reliability of early warning information and the reliability of the system.
The invention also provides a mine safety situation collaborative awareness method based on deduction learning, which comprises the following steps:
step one: firstly, acquiring sensing data by using sensors arranged in an environment and a patrol robot in a mine safety production process through a data acquisition module, and constructing a sensing data set for forming mine safety situation grade evaluation;
secondly, denoising, normalizing and outlier processing are carried out on the perception data set through a data processing module, influence factors of coal mine safety production are analyzed, meanwhile, characteristic parameters for evaluating mine safety situation level are constructed through steps of knowledge extraction, fusion and processing in combination with cause analysis of typical accidents, and unlabeled data are collected to form a training data set;
step two: the method comprises the steps of collecting and judging related data of mine safety situation levels by combining mine typical safety accident cause analysis, mine safety production standards and mine production whole processes, and carrying out abstract expression in a first-order predicate logic form to form an expert knowledge base; dividing the coal mine safety situation level into five levels of low risk, general risk, medium risk, important risk and extra-large risk by utilizing a coal mine safety situation logic reasoning strategy based on a fuzzy logic algorithm;
step three: taking a small part of the data set in the first step as S for performing supervised training on the mine safety situation level evaluation pre-training model to form an initial classifier C, and remaining unmarked data sets as S';
step four: inputting the data in the S' data set into an initial classifier C for prediction to obtain a pseudo tag of each data;
step five: comparing the pseudo tag of each data obtained in the step four with the result output by the logic reasoning module, minimizing inconsistency, obtaining an anti-deduction tag of the data through an anti-deduction learning framework, adding the data and the anti-deduction tag into the data set S and correspondingly replacing the data set S to form a new data set S;
step six: the new data set S obtained in the step five is used for training the classifier C, model parameters are updated to form a new classifier C, the new data set S is input into the classifier C for prediction, and the step five is repeated;
step seven: and (3) continuously cycling the process of the step six until the inconsistency is not required to be minimized or the machine learning model is converged, and after the cycling is stopped, constructing a final mine safety situation grade assessment model.
As a preferable mode, the mine safety situation grade assessment model of the machine learning module in the third step adopts a decision layer fusion method in a multi-mode fusion strategy, and the logic reasoning module directly uses the data set constructed in the first step.
As a preferred mode, constructing an expert knowledge base of the logic reasoning module in the second step, combining the analysis of typical mine safety accident causes, mine safety production standards and mine production complete flow, collecting and judging the relevant basis of the mine safety situation level, and carrying out abstract expression in a first-order predicate logic mode to form the expert knowledge base;
the method comprises the following specific steps:
a1, introducing a neural network as a predicate in a logic formula, combining deep learning with soft and hard logic constraint, and allowing a plurality of neural networks to be used and fully incorporating constraint and relation information; here, each logical rule statement is mapped to a real value by a mapping function O:
O(y,x,gnn(xnn,wnn))->R；
wherein x is an observation variable in a logic statement, y is a target variable, gnn is a neural network, and xnn and wnn are input and model parameters of the neural network respectively;
a2, setting a weight w for each logic rule statement, and finally, enabling an energy function E=w T [O(y,x,gnn(xnn,wnn))]->R；
Wherein w is T Is the transpose of the weight component vector of each logical statement, [ O (y, x, gnn (xnn, wnn))]Is a vector formed by mapping real values of each logic statement;
a3, dividing the coal mine security situation level into the intervals of real values of the energy function E: low risk, general risk, medium risk, significant risk, and extra-high risk.
As a preference, the policy procedure for minimizing inconsistencies in step five is as follows:
the method comprises the following specific steps:
s1, comparing a prediction result of a neural network in a machine learning module with a logic reasoning result of a logic reasoning module;
s2, setting a threshold value, generating a pseudo tag based on a prediction result of the neural network in the machine learning module, if the difference value between the prediction result of the neural network and the logic reasoning result does not exceed the threshold value, reserving the prediction result of the neural network, otherwise, replacing the pseudo tag with the logic reasoning result;
and S3, taking the tag obtained in the step S2 as an anti-deduction tag, adding the anti-deduction tag and corresponding data into a data set, and correspondingly replacing the anti-deduction tag and the corresponding data, and retraining the mine safety situation grade evaluation pre-training model.
In the invention, because the environment under the mine is limited, the acquired original perception data has larger noise, the data of the data acquisition module is firstly subjected to denoising, normalization and outlier processing by the data processing module, and then the kernel characteristics for determining the data difference are extracted from the original high-dimensional data by utilizing the variation self-encoder, so that the noise can be effectively reduced and removed, and the evaluation of the mine safety situation level is facilitated to be carried out rapidly and accurately. And combining case analysis and expert priori knowledge to construct a knowledge base for mine safety situation level assessment, and analyzing and reasoning the knowledge base and historical data by using a fuzzy algorithm so as to obtain a more accurate assessment structure. The method is characterized in that part of the data set is manually marked for fine adjustment of a machine learning pre-training model, and then logic reasoning is realized through a fuzzy logic algorithm and a knowledge base for security situation level assessment, so that the constructed data-driven and knowledge-driven mine security situation prediction method comprehensively considers the logic reasoning and machine learning characteristics, and finally a mine security situation level assessment model based on anti-deduction learning can be obtained through modeling the credibility of the two. The method is different from the traditional machine learning model prediction method, combines machine learning and logic reasoning, comprehensively considers the prediction results of the machine learning model and the logic reasoning, and realizes accurate assessment of the mine safety situation level.
Drawings
FIG. 1 is a mine safety situation collaborative awareness mechanism construction based on deductive learning in the invention;
FIG. 2 is a block diagram of a mine safety situation collaborative awareness system based on deductive learning;
FIG. 3 is a general flow chart of a mine safety situation collaborative awareness method based on deductive learning;
fig. 4 is a flow chart of a portion of the combination of machine learning and logical reasoning in accordance with the present invention.
Detailed Description
As shown in fig. 1 and 2, the invention provides a mine safety situation collaborative awareness system based on deduction learning, which comprises a data acquisition module, a data processing module, a machine learning module, a logic reasoning module and a machine learning and logic reasoning fusion module;
the data acquisition module is connected with the data processing module and is used for constructing a perception data set for mine safety situation grade assessment according to perception data acquired by the environment and the sensors in the inspection robot in the mine safety production process and sending the perception data set to the data processing module;
the data processing module is respectively connected with the machine learning module and the logic reasoning module, is used for denoising, normalizing and outlier processing of the received perception data set, is used for determining characteristic parameters for evaluating the mine safety situation level by combining expert experience and priori knowledge with a knowledge base, converts an original perception data set into a group of core characteristic data sets with obvious statistical significance by automatically constructing new characteristics so as to improve the robustness of the model, and finally is respectively sent to the machine learning module and the logic reasoning module;
the machine learning module is used for inputting processed perception data into the sub-neural networks of different modes, integrating and fusing the prediction results of the sub-neural networks of all modes by adopting a decision layer fusion method in a multi-mode fusion strategy to obtain a final overall decision, inputting the overall decision into a mine safety situation level assessment pre-training model, finally constructing a data-driven safety situation perception model, generating a data-driven mine safety situation level assessment result according to training characteristics, and finally transmitting the mine safety situation level assessment result to the machine learning and logic reasoning fusion module;
the decision layer fusion method comprises the following steps: the method can fully utilize information of each mode, improve the fusion effect and system performance, and simultaneously has the advantages of flexibility, expandability, interpretability and the like, so that the processing of multi-mode tasks is more effective and reliable.
The logic reasoning module is used for determining the basis for evaluating the mine safety situation level according to expert experience and priori knowledge, carrying out abstract expression in a first-order predicate logic form to form an expert knowledge base, constructing a knowledge-driven safety situation logic reasoning model based on analysis and reasoning of the knowledge base and historical data by a fuzzy logic algorithm, generating a knowledge-driven mine safety situation level evaluation result according to training characteristics, and sending the mine safety situation level evaluation result to the machine learning and logic reasoning fusion module;
the machine learning and logic reasoning fusion module is respectively connected with the machine learning module and the logic reasoning module, is used for comprehensively considering logic reasoning and machine learning characteristics, modeling the credibility of the machine learning module and the logic reasoning fusion module, generating a pseudo tag based on a prediction result of a neural network in the machine learning module, correcting the pseudo tag by utilizing the result of the logic reasoning module, updating a mine safety situation level evaluation pre-training model, and finally constructing and forming a mine safety situation collaborative sensing system based on anti-deduction learning.
The flow of machine learning in combination with logical reasoning is shown in fig. 4. The logic reasoning module compares the knowledge base with the evaluation results of the machine learning module through analysis and reasoning of the historical data, so that the most likely evaluation results are obtained, the results are returned to the security situation evaluation model, model parameters are updated, and a final mine security situation grade evaluation model is obtained, so that the final mine security situation grade evaluation results are obtained. For the misjudgment or missed judgment situation obtained by the machine learning module, the logic reasoning module can correct the misjudgment or missed judgment situation by reasoning and deducing according to rules and semantics of the knowledge base, so that the judgment and classification accuracy of the model is improved. In addition, in the reasoning process, the logic reasoning module can also find unknown threat factors and update relevant safety information in the knowledge base so as to better evaluate the safety threat possibly occurring in the future.
Preferably, the mine safety situation level assessment pre-training model adopts a variation automatic encoder. Because the environment under the mine is limited, the acquired data has larger noise, and the core features for determining the data difference are extracted from the original high-dimensional data by using an automatic encoder, so that the noise can be effectively reduced and removed, and the generalization capability of the model is obviously improved.
In the invention, because the logic reasoning module is based on the fuzzy logic algorithm, different from the traditional Boolean logic algorithm, the rule used by the fuzzy logic algorithm is more flexible, and the ambiguity concept is mapped to the real number interval to calculate, so that the uncertainty of different degrees can be processed. The method also has certain processing capacity for data containing certain noise or inaccuracy, and can improve the robustness of the algorithm. Meanwhile, the fuzzy logic algorithm can handle a large number of uncertain or fuzzy situations and can be adjusted according to the needs, so that the sensing system is very flexible. In addition, the logic reasoning module is arranged and the machine learning module is also arranged, so that the advantages of the logic reasoning module and the machine learning module can be fully combined, the advantages of asynchronous data can be processed by utilizing a multi-mode late fusion method, and meanwhile, the logic reasoning module directly uses the original data to make up for the defect that the original data neglects low-level characteristics among a plurality of modes. Through the arrangement of the machine learning and logic reasoning fusion module, an effectively fused knowledge reasoning and data-driven mine safety situation collaborative sensing system can be formed, so that the characteristics of logic reasoning and machine learning can be fully and comprehensively considered, the evaluation result of the machine learning module can be detected and verified by utilizing the logic reasoning module, the mine safety situation level evaluation pre-training model can be updated, and further the mine safety situation collaborative sensing system based on anti-deduction learning can be formed, the safety level of a coal mine can be rapidly and accurately evaluated, and the safety production operation of the coal mine can be guaranteed. Because of the complex and varying mine environments, it may be difficult to use other machine learning methods to arrive at the correct conclusion with low reliability or low data volume. The system builds a knowledge-driven safety situation logic reasoning model through a fuzzy logic algorithm, builds a data-driven safety situation sensing model through machine learning algorithms such as a variation automatic encoder and the like, and based on anti-deduction learning, combines machine learning and logic reasoning prediction results, and updates a mine safety situation level assessment model after comprehensive consideration to obtain a final mine safety situation level assessment model, so that a correct conclusion can be obtained through deduction from a large amount of known information, and the system can be rapidly adapted to the uncertain and changed fields. The invention has simple structure, strong universality and easy implementation, effectively combines machine learning and logic reasoning together in a balanced and reciprocal mode, is beneficial to correcting perceived errors and enhances the robustness of the system. Aiming at the complex mine environment, the system can realize accurate assessment of the security situation level and can improve the reliability of early warning information and the reliability of the system.
As shown in fig. 3, the invention also provides a mine safety situation collaborative awareness method based on deduction learning, which comprises the following steps:
step one: firstly, acquiring sensing data by using sensors arranged in an environment and a patrol robot in a mine safety production process through a data acquisition module, and constructing a sensing data set for forming mine safety situation grade evaluation;
secondly, denoising, normalizing and outlier processing are carried out on the perception data set through a data processing module, influence factors of coal mine safety production are analyzed, meanwhile, characteristic parameters for evaluating mine safety situation level, such as gas concentration, wind flow, dust concentration, coal and rock mass temperature, personnel unsafe behaviors and the like, are constructed through steps of knowledge extraction, fusion and processing in combination with cause analysis of typical accidents, and unlabeled data are collected to form a training data set;
step two: the method comprises the steps of collecting and judging related data of mine safety situation levels by combining mine typical safety accident cause analysis, mine safety production standards and mine production whole processes, and carrying out abstract expression in a first-order predicate logic form to form an expert knowledge base; dividing the coal mine safety situation level into five levels of low risk, general risk, medium risk, important risk and extra-large risk by utilizing a coal mine safety situation logic reasoning strategy based on a fuzzy logic algorithm;
step three: taking a small part of the data set in the first step as S for performing supervised training on the mine safety situation level evaluation pre-training model to form an initial classifier C, and remaining unmarked data sets as S';
the mine safety situation level evaluation pre-training model adopts a variation automatic encoder, and the loss function of the variation automatic encoder is shown in the following formula (1)
The following is shown:
where j is the dimension of the hidden variable, μ and σ i Is the mean and variance of the variational probability distribution of the hidden variable.
Step four: inputting the data in the S' data set into an initial classifier C for prediction to obtain a pseudo tag of each data;
step five: comparing the pseudo tag of each data obtained in the step four with the result output by the logic reasoning module, minimizing inconsistency, obtaining an anti-deduction tag of the data through an anti-deduction learning framework, adding the data and the anti-deduction tag into the data set S and correspondingly replacing the data set S to form a new data set S;
this step is exemplified as follows:
assuming that a prediction result output by the mine safety situation level assessment pre-training model, namely a pseudo tag is a medium risk, and the output result of the logic reasoning module is a general risk, a medium risk or a major risk, the security situation level predicted by the neural network is regarded as being adjacent to the logic reasoning, and the tag is reserved; if the output result is low risk or extra risk, the output result is not adjacent, and in order to minimize the inconsistency, the pseudo tag is corrected to be a logical reasoning result according to the logical reasoning result. Finally, the tag is added to the dataset as an deduction tag and its corresponding data.
Step six: the new data set S obtained in the step five is used for training the classifier C, model parameters are updated to form a new classifier C, the new data set S is input into the classifier C for prediction, and the step five is repeated;
step seven: and (3) continuously cycling the process of the step six until the inconsistency is not required to be minimized or the machine learning model is converged, and after the cycling is stopped, constructing a final mine safety situation grade assessment model.
As a preferable mode, the mine safety situation grade assessment model of the machine learning module in the third step adopts a decision layer fusion method in a multi-mode fusion strategy, and the logic reasoning module directly uses the data set constructed in the first step.
As a preferred mode, constructing an expert knowledge base of the logic reasoning module in the second step, combining the analysis of typical mine safety accident causes, mine safety production standards and mine production complete flow, collecting and judging the relevant basis of the mine safety situation level, and carrying out abstract expression in a first-order predicate logic mode to form the expert knowledge base;
the method comprises the following specific steps:
a1, introducing a neural network as a predicate in a logic formula, combining deep learning with soft and hard logic constraint, and allowing a plurality of neural networks to be used and fully incorporating constraint and relation information; here, each logical rule statement is mapped to a real value by a mapping function O:
O(y,x,gnn(xnn,wnn))->R；
wherein x is an observation variable in a logic statement, y is a target variable, gnn is a neural network, and xnn and wnn are input and model parameters of the neural network respectively;
a2, setting a weight w for each logic rule statement, and finally, enabling an energy function E=w T [O(y,x,gnn(xnn,wnn))]->R；
Wherein w is T Is the transpose of the weight component vector of each logical statement, [ O (y, x, gnn (xnn, wnn))]Is a vector formed by mapping real values of each logic statement;
a3, dividing the coal mine security situation level into the intervals of real values of the energy function E: low risk, general risk, medium risk, significant risk, and extra-high risk.
As a preference, the policy procedure for minimizing inconsistencies in step five is as follows:
comparing the pseudo tag predicted by the neural network with a result obtained by logic reasoning, if the security situation level predicted by the neural network is adjacent to the logic reasoning, reserving the tag, if the security situation level predicted by the neural network is not adjacent to the logic reasoning, taking the logic reasoning result as the reference, correcting the pseudo tag into the logic reasoning result, and finally adding the tag as an anti-deduction tag and corresponding data thereof into a data set for corresponding replacement;
the method comprises the following specific steps:
s1, comparing a prediction result of a neural network in a machine learning module with a logic reasoning result of a logic reasoning module;
s2, setting a threshold value, generating a pseudo tag based on a prediction result of the neural network in the machine learning module, if the difference value between the prediction result of the neural network and the logic reasoning result does not exceed the threshold value, reserving the prediction result of the neural network, otherwise, replacing the pseudo tag with the logic reasoning result;
and S3, taking the tag obtained in the step S2 as an anti-deduction tag, adding the anti-deduction tag and corresponding data into a data set, and correspondingly replacing the anti-deduction tag and the corresponding data, and retraining the mine safety situation grade evaluation pre-training model.
In the invention, because the environment under the mine is limited, the acquired original perception data has larger noise, the data of the data acquisition module is firstly subjected to denoising, normalization and outlier processing by the data processing module, then the kernel characteristics for determining the data difference are extracted from the original high-dimensional data by the variable self-encoder, and a data set for training of a machine learning model is constructed, so that the noise can be effectively reduced and removed, and the mine safety situation level can be rapidly and accurately evaluated. And combining case analysis and expert priori knowledge to construct a knowledge base for evaluating the mine safety situation level. The knowledge-driven security situation logic reasoning model is designed by utilizing a fuzzy logic algorithm, the data-driven security situation sensing model is designed by adopting a machine learning algorithm such as a variation encoder and the like, the logic reasoning and machine learning characteristics are comprehensively considered, and the reliability of the logic reasoning and the machine learning characteristics is modeled, so that the mine security situation grade assessment model based on anti-deduction learning can be finally obtained. The method is different from the traditional machine learning model prediction method, combines machine learning and logic reasoning, comprehensively considers the prediction results of the machine learning model and the logic reasoning, and realizes accurate assessment of the mine safety situation level.
Claims (4)
1. The mine safety situation collaborative awareness system based on anti-deduction learning is characterized by comprising a data acquisition module, a data processing module, a machine learning module, a logic reasoning module and a machine learning and logic reasoning fusion module:
the data acquisition module is connected with the data processing module and is used for constructing a perception data set for mine safety situation grade assessment according to perception data acquired by the environment and the sensors in the inspection robot in the mine safety production process and sending the perception data set to the data processing module;
the data processing module is respectively connected with the machine learning module and the logic reasoning module, is used for determining characteristic parameters for evaluating the mine safety situation level by combining expert experience and priori knowledge with a knowledge base, converts an original perception data set into a group of core characteristic data sets with obvious statistical significance by automatically constructing new characteristics, and respectively sends the core characteristic data sets to the machine learning module and the logic reasoning module;
the machine learning module is used for inputting the processed perception data into the sub-neural networks of different modes, integrating and fusing the prediction results of the sub-neural networks of each mode to obtain a final overall decision, inputting the overall decision into the mine safety situation level assessment pre-training model, finally constructing a data-driven safety situation perception model, generating a data-driven mine safety situation level assessment result according to training characteristics, and finally transmitting the mine safety situation level assessment result to the machine learning and logic reasoning fusion module;
the logic reasoning module is used for determining the basis for evaluating the mine safety situation level according to expert experience and priori knowledge, carrying out abstract expression in a first-order predicate logic form to form an expert knowledge base, constructing a knowledge-driven safety situation logic reasoning model based on analysis and reasoning of the knowledge base and historical data by a fuzzy logic algorithm, generating a knowledge-driven mine safety situation level evaluation result according to training characteristics, and sending the mine safety situation level evaluation result to the machine learning and logic reasoning fusion module;
the machine learning and logic reasoning fusion module is respectively connected with the machine learning module and the logic reasoning module, is used for comprehensively considering logic reasoning and machine learning characteristics, modeling the credibility of the machine learning module and the logic reasoning fusion module, generating a pseudo tag based on a prediction result of a neural network in the machine learning module, correcting the pseudo tag by utilizing the result of the logic reasoning module, updating a mine safety situation level evaluation pre-training model, and finally constructing and forming a mine safety situation collaborative sensing system based on anti-deduction learning.
2. The mine safety situation collaborative awareness system based on deductive learning according to claim 1, wherein the mine safety situation level assessment pre-training model employs a variational automatic encoder.
3. The mine safety situation collaborative sensing method based on anti-deduction learning is characterized by comprising the following steps of:
step one: firstly, acquiring sensing data by using sensors arranged in an environment and a patrol robot in a mine safety production process through a data acquisition module, and constructing a sensing data set for forming mine safety situation grade evaluation;
secondly, denoising, normalizing and outlier processing are carried out on the perception data set through a data processing module, influence factors of coal mine safety production are analyzed, meanwhile, characteristic parameters for evaluating mine safety situation level are constructed through steps of knowledge extraction, fusion and processing in combination with cause analysis of typical accidents, and unlabeled data are collected to form a training data set;
step two: the method comprises the steps of collecting and judging related data of mine safety situation levels by combining mine typical safety accident cause analysis, mine safety production standards and mine production whole processes, and carrying out abstract expression in a first-order predicate logic form to form an expert knowledge base; dividing the coal mine safety situation level into five levels of low risk, general risk, medium risk, important risk and extra-large risk by utilizing a coal mine safety situation logic reasoning strategy based on a fuzzy logic algorithm;
step three: taking a small part of the data set in the first step as S for performing supervised training on the mine safety situation level evaluation pre-training model to form an initial classifier C, and remaining unmarked data sets as S';
step four: inputting the data in the S' data set into an initial classifier C for prediction to obtain a pseudo tag of each data;
step five: comparing the pseudo tag of each data obtained in the step four with the result output by the logic reasoning module, minimizing inconsistency, obtaining an anti-deduction tag of the data through an anti-deduction learning framework, adding the data and the anti-deduction tag into the data set S and correspondingly replacing the data set S to form a new data set S;
step six: the new data set S obtained in the step five is used for training the classifier C, model parameters are updated to form a new classifier C, the new data set S is input into the classifier C for prediction, and the step five is repeated;
step seven: and (3) continuously cycling the process of the step six until the inconsistency is not required to be minimized or the machine learning model is converged, and after the cycling is stopped, constructing a final mine safety situation grade assessment model.
4. A mine safety situation collaborative awareness method based on deductive learning according to claim 3, characterized in that the policy procedure of minimizing inconsistencies in step five is as follows:
the method comprises the following specific steps:
s1, comparing a prediction result of a neural network in a machine learning module with a logic reasoning result of a logic reasoning module;
s2, setting a threshold value, generating a pseudo tag based on a prediction result of the neural network in the machine learning module, if the difference value between the prediction result of the neural network and the logic reasoning result does not exceed the threshold value, reserving the prediction result of the neural network, otherwise, replacing the pseudo tag with the logic reasoning result;
and S3, taking the tag obtained in the step S2 as an anti-deduction tag, adding the anti-deduction tag and corresponding data into a data set, and correspondingly replacing the anti-deduction tag and the corresponding data, and retraining the mine safety situation grade evaluation pre-training model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311064467.8A CN117035424A (en) | 2023-08-23 | 2023-08-23 | Mine safety situation collaborative sensing system and method based on anti-deduction learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311064467.8A CN117035424A (en) | 2023-08-23 | 2023-08-23 | Mine safety situation collaborative sensing system and method based on anti-deduction learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117035424A true CN117035424A (en) | 2023-11-10 |
Family
ID=88626207
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311064467.8A Pending CN117035424A (en) | 2023-08-23 | 2023-08-23 | Mine safety situation collaborative sensing system and method based on anti-deduction learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117035424A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117892212A (en) * | 2024-03-14 | 2024-04-16 | 国网上海市电力公司 | Distributed heterogeneous energy station situation awareness monitoring method |
-
2023
- 2023-08-23 CN CN202311064467.8A patent/CN117035424A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117892212A (en) * | 2024-03-14 | 2024-04-16 | 国网上海市电力公司 | Distributed heterogeneous energy station situation awareness monitoring method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110633855B (en) | Bridge health state detection and management decision making system and method | |
| US20180336437A1 (en) | Streaming graph display system with anomaly detection | |
| CN111585997A (en) | Network flow abnormity detection method based on small amount of labeled data | |
| CN112529341B (en) | Drilling well leakage probability prediction method based on naive Bayesian algorithm | |
| Du et al. | GAN-based anomaly detection for multivariate time series using polluted training set | |
| US11662718B2 (en) | Method for setting model threshold of facility monitoring system | |
| WO2023142424A1 (en) | Power financial service risk control method and system based on gru-lstm neural network | |
| CN117035424A (en) | Mine safety situation collaborative sensing system and method based on anti-deduction learning | |
| Dong | A tutorial on nonlinear time-series data mining in engineering asset health and reliability prediction: concepts, models, and algorithms | |
| CN112966714A (en) | Edge time sequence data anomaly detection and network programmable control method | |
| CN116341901B (en) | Integrated evaluation method for landslide surface domain-monomer hazard early warning | |
| CN111126820A (en) | Electricity stealing prevention method and system | |
| CN114519923A (en) | Intelligent diagnosis and early warning method and system for power plant | |
| Zhang et al. | A real-time anomaly detection algorithm/or water quality data using dual time-moving windows | |
| CN110580213A (en) | Database anomaly detection method based on cyclic marking time point process | |
| CN115964503B (en) | Safety risk prediction method and system based on community equipment facilities | |
| CN116738192A (en) | Digital twinning-based security data evaluation method and system | |
| CN117273440A (en) | Engineering construction Internet of things monitoring and managing system and method based on deep learning | |
| Liu et al. | Residual useful life prognosis of equipment based on modified hidden semi-Markov model with a co-evolutional optimization method | |
| Pattnaik et al. | A survey on machine learning techniques used for software quality prediction | |
| Li et al. | Information granularity with the self-emergence mechanism for event detection in WSN-based tunnel health monitoring | |
| CN113328881A (en) | Topology sensing method, device and system for non-cooperative wireless network | |
| Ben Khedher et al. | Statistical quality control based on control charts and process efficiency index by the application of fuzzy approach (case study: Ha'il, Saudi Arabia) | |
| CN112819264A (en) | Weighted random mixed semantic method for battle effectiveness evaluation | |
| Li et al. | A decision system based on intelligent perception and decision for scene ventilation safety |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |