Management method and device for aircraft refueling operation

Abstract

The invention discloses a management method and device for aircraft refueling operation. The management method comprises the following steps: constructing a knowledge graph about aircraft refueling operation; constructing a template library by utilizing the knowledge graph; constructing a plurality of agents with different functions according to the knowledge graph and the template library; constructing a detection rule for judging a risk scene according to natural language input of a user; obtaining a training set of the intelligent agent; training the intelligent agent by using the training set and the detection rule; managing an aircraft refueling operation area by using the trained intelligent agent, and making a corresponding management strategy according to input information; according to the management scheme, the detection rule is defined by using a natural language, the utilization rate of physical defense and civil defense can be improved to the maximum extent, known risks can be effectively prevented, and the remaining risks are found through anomaly recognition, so that the management system is continuously improved.

Description

Technical Field

[0001] The present invention relates to the field of civil aviation management, and in particular to a management method and device for aircraft refueling operations. Background Art

[0002] Aircraft refueling operation is one of the most important links in aviation safety. In order to ensure the safety and accuracy of refueling operation, it is necessary to introduce advanced technical means to monitor and control various risk factors in the operation process. Traditional refueling operation has potential risks such as human error and equipment failure. Therefore, the present invention aims to provide a new aircraft refueling safety integrated protection system to improve the safety and efficiency of refueling operation.

[0003] The existing intelligent fire warning system needs to predefine certain abnormal scenarios. Identifying abnormalities requires a large amount of labeled abnormal data. The huge workload of data collection and labeling is often not feasible in the short term, and the project is difficult to implement. An even more difficult obstacle is that there are many states in which abnormalities occur, which are difficult to list and simulate accurately. This makes it difficult for the existing system to identify new safety abnormal events, and the warnings for known risks are inaccurate. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a management solution for aircraft refueling operations with high risk identification accuracy and small training set requirement.

[0005] In order to achieve the above-mentioned objectives, on the one hand, the present invention provides a management method for aircraft refueling operations, including: constructing a knowledge graph about aircraft refueling operations; constructing a template library using the knowledge graph; constructing multiple intelligent agents with different functions based on the knowledge graph and the template library; constructing detection rules for judging risk scenarios based on the user's natural language input; obtaining a training set for the intelligent agent; the training set includes one or more of pictures, videos and sounds; training the intelligent agent using the training set and the detection rules; managing the aircraft refueling operation area using the trained intelligent agent, and making corresponding management strategies based on the input information; the input information includes the user's natural language input or one or more of pictures, videos and sounds of the aircraft refueling operation area.

[0006] Optionally, the construction of a knowledge graph about aircraft refueling operations includes: determining a keyword list for knowledge terminal nodes; establishing strongly associated knowledge point segment slices based on the keyword list; and constructing a cross-knowledge point knowledge graph based on the keyword list and the strongly associated knowledge point segment slices.

[0007] Optionally, determining a keyword list for knowledge terminal nodes includes: collecting and maintaining a keyword list covering all important concepts and terms in the field; updating and expanding the keyword list using natural language processing tools; determining formal definitions of keywords and contexts to form a keyword list.

[0008] Optionally, the step of establishing strongly associated knowledge point segment slices based on the keyword list includes: analyzing a text corpus to extract combinations of knowledge points that frequently appear together; using topic modeling technology to identify and separate segments, and establish associations between the knowledge point combinations; establishing a segment index and marking strongly associated knowledge points for rapid retrieval and integration.

[0009] Optionally, the use of the knowledge graph to construct a template library includes: collecting common questions and answer patterns in the industry, and using the common questions and answer patterns to construct a standardized template library; using natural language generation technology to generate answers that conform to the templates; combining the template library with the knowledge graph, and dynamically selecting and filling the most suitable templates according to the input questions, thereby forming a final template library.

[0010] Optionally, the determination of detection rules for judging risk scenarios based on the user's natural language input includes: determining detection rules or paradigms based on the user's natural language input; re-confirming the detection rules or paradigms through similarity identification technology; testing the accuracy of the detection rules or paradigms using a preset test set and providing feedback; wherein the preset test set contains verified detection rules or paradigms.

[0011] Optionally, determining the detection rules or paradigm based on the user's natural language input includes: constructing a rule language or paradigm to allow the user's natural language input to define the detection rules; determining key prompt words and / or phrases, which serve as trigger conditions for establishing and modifying the detection rules; and using natural language processing technology to parse the natural language rules provided by the user.

[0012] Optionally, the use of a preset test set to test the accuracy of the detection rule or paradigm and provide feedback includes: constructing a preset test set that includes multiple scenarios and situations; writing test cases and expected outputs, wherein the test cases and expected outputs cover different types of detection rules or paradigms; using an automated testing framework to execute the detection rule or paradigm, and comparing the execution results with the preset test set, thereby evaluating the accuracy and effectiveness of the detection rule or paradigm.

[0013] On the other hand, the present invention also provides a management device for aircraft refueling operations, including: a first construction unit, used to construct a knowledge graph about aircraft refueling operations; a second construction unit, used to construct a template library using the knowledge graph; a third construction unit, used to construct multiple intelligent agents with different functions based on the knowledge graph and the template library; a fourth construction unit, used to construct detection rules for judging risk scenarios based on the user's natural language input; an acquisition unit, used to acquire a training set of the intelligent agent; the training set includes one or more of pictures, videos and sounds; a training unit, used to train the intelligent agent using the training set and the detection rules; an output unit, used to manage the aircraft refueling operation area using the trained intelligent agent, and make corresponding management strategies based on the input information; the input information includes the user's natural language input or one or more of pictures, videos and sounds of the aircraft refueling operation area.

[0014] The beneficial effects of the present invention compared to the prior art are: the management solution ensures that the big model can correctly provide information in the field of aircraft refueling operations through the professional knowledge base and knowledge graph. In addition, by using natural language to define detection rules, the utilization rate of physical defense and human defense can be maximized. Finally, by using computer vision and big models to achieve multimodal recognition of risk scenarios, known risks can be effectively prevented, and legacy risks can be discovered through anomaly recognition, and management systems can be continuously improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of a management method of aircraft refueling operation provided by one embodiment of the present invention; Figure 2 is a flow chart for constructing a knowledge graph about aircraft refueling operations provided by an embodiment of the present invention; Figure 3 It is a flowchart of constructing a detection rule for determining a risk scenario according to a user's natural language input provided by an embodiment of the invention; Figure 4 It is a structural diagram of a management device for aircraft refueling operations provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] Reference Figure 1This embodiment provides a method for managing aircraft refueling operations, comprising the following steps: Step S10: Construct a knowledge graph about aircraft refueling operations.

[0018] Reference Figure 2 In this embodiment, step S10 includes: Step S101: Determine a keyword list for knowledge terminal nodes.

[0019] Specifically, first, domain experts are involved in collecting and maintaining a keyword list that covers all important concepts and terms in the field. Then, natural language processing tools such as part-of-speech tagging and entity recognition are used to automatically update and expand the keyword list. Finally, formal definitions and contexts of keywords are determined to ensure accuracy and consistency.

[0020] In this embodiment, the natural language processing tool adopts NLTK (Natural Language Toolkit). It should be noted that the protection scope of the present invention is not limited to this. Other natural language processing tools, such as Spacy, Gensim, Stanford CoreNLP, WordNet, TextBlob, BERT, AllenNLP, etc. are all within the protection scope of the present invention.

[0021] Step S102: creating strongly associated knowledge point paragraph slices according to the keyword list.

[0022] Specifically, first, analyze the text corpus and extract the knowledge points that often appear together. Then, use topic modeling techniques (such as LDA) to identify and separate the segments and establish the associations between them. Finally, create a segment index and mark the strongly associated knowledge points for fast retrieval and integration.

[0023] Step S103: construct a knowledge graph across knowledge points based on the keyword list and strongly associated knowledge point segment slices.

[0024] Specifically, when the keyword list and the strongly associated knowledge point segment slicing are completed, the knowledge graph can be constructed. Since the construction of the knowledge graph belongs to the conventional technical means in this field, the specific construction process will not be repeated in this embodiment.

[0025] It should be noted that in this embodiment, when constructing the knowledge graph, graph database technology is used to store and manage the relationships between different knowledge points, so that the associations of long-span complex structures can be captured, and efficient query and reasoning mechanisms can be implemented, spanning multiple knowledge points to solve complex problems.

[0026] The construction process of the knowledge graph recorded in this embodiment processes the keyword list through natural language processing tools, which can fully expand the keyword list, greatly reduce the content required for knowledge graph construction, make up for the shortcomings of less knowledge materials in the field of aircraft refueling operations, and greatly reduce the time of data processing. In addition, by using topic modeling technology to establish strongly associated knowledge point paragraph slices, the accuracy of the knowledge graph is high.

[0027] Step S20: construct a template library using the knowledge graph.

[0028] Specifically, first, collect common questions and answer patterns in the industry and build a standardized template library.

[0029] Next, natural language generation (NLG) technology is used to automatically generate answers that conform to the template to ensure the accuracy and consistency of the content.

[0030] Finally, the template library is combined with the knowledge graph to dynamically select and fill in the most suitable template based on the input question.

[0031] It should be noted that the template library recorded in this embodiment is pre-defined according to the usage scenario, and the appropriate judgment criteria are determined by the "usage scenario". For example, if the user asks to "write a network security report", the system will extract the network report template from the template library. This extraction process mainly involves feature annotation of user questions and comparison with manually annotated feature questions to standardize user questions. When the question is classified as a security report, the system will take out the template library stored in the knowledge graph. The two are combined, not one after the other.

[0032] In some other embodiments, the management method also establishes a prompt word project for accuracy judgment to reject uncertain questions.

[0033] Specifically, we first build an accuracy assessment model, which determines the validity and weight of prompt words based on statistical analysis and machine learning techniques. Then, we use this accuracy assessment model to identify and mark keywords and phrases that are uncertain and ambiguous. Users can also set thresholds and logical rules to provide reasonable rejection or improvement feedback for questions identified as uncertain.

[0034] By using accuracy judgment prompts to project accuracy judgment and feedback, detect and handle uncertain problems, the accuracy and credibility of the output information can be ensured.

[0035] Step S30: Construct multiple intelligent agents with different functions based on the knowledge graph and the template library.

[0036] In this embodiment, there are multiple intelligent agents, each of which is responsible for different detection tasks, such as target recognition, anomaly detection, etc.

[0037] It should be noted that the intelligent agent used to achieve target recognition in this embodiment includes multiple identification algorithms, because target recognition is not a target in a specific industrial scenario, but a series of time-ordered action recognition. For example, if we recognize an action of connecting an oil pipe, we need to recognize that the worker extracts the oil pipe, climbs up the ladder, connects the oil pipe, and then climbs down the ladder. The recognition of different actions (targets) requires separate training, and it is a set. The intelligent agent is the business agent that operates the entire inspection action. It is responsible for connecting the entire inspection process in series, and through a workflow loop inspection, it completes multi-modal state recognition based on the output of the target recognition algorithm and other system inputs to ensure that personnel and objects are in a safe state.

[0038] In this embodiment, the algorithm adopted by the intelligent agent is not limited. For example, convolutional neural network (CNN) and recurrent neural network (RNN) in deep learning can all achieve the technical effect to be achieved in this embodiment.

[0039] In this embodiment, the intelligent agent of anomaly detection uses a clustering algorithm and a time series data processing model, wherein the clustering algorithm includes but is not limited to k-means, DBSCAN, etc., and the time series data processing model includes but is not limited to a long short-term memory network (LSTM), etc.

[0040] Step S40: Constructing detection rules for determining risk scenarios based on the user's natural language input.

[0041] Reference Figure 3 In this embodiment, step S40 includes: Step S401: determining a detection rule or paradigm according to a natural language input of a user.

[0042] Specifically, first, design and establish a rule language or paradigm that allows natural language input to define detection rules. Then determine key prompt words and phrases as trigger conditions for rule establishment and modification. Finally, use natural language processing (NLP) technology, including grammatical parsing and semantic analysis, to parse the natural language rules provided by users.

[0043] The following is an example of an alarm judgment scenario: The Large Language Model (LLM) outputs judgments based on the regular expression a&&b&&c and the alarm content.

[0044] If it is false, please answer: The alarm exists, please go to the alarm label immediately to check.

[0045] If true, please answer: The current alarm does not exist, please pay attention.

[0046] Give the alarm content: a: no alarm in the hall = false; b: no alarm in the factory area = true; c: no alarm in the dormitory = false; or: The hall alarm status is a set, and each value is composed of "sensor number: Boolean value". For example, sensor No. 3 is expressed as 3:true. When there are 4 sensors in total, the overall expression is {1:true, 2:false; 3:true; 4:true}.

[0047] If all the values ​​in the set are true, answer: No alarm in the lobby = true.

[0048] If false appears in the set, answer: no alarm in the lobby = false.

[0049] Give an answer based on the following lobby alarm situation: lobby alarm situation: {1:true,2:false;3:true;4:true}.

[0050] Step S402: re-confirm the detection rule or paradigm by using similarity identification technology.

[0051] Specifically, first, a similarity calculation algorithm (such as the cosine similarity algorithm) is used to reconfirm and verify the natural language rules. Then, the natural language rules are converted into a standardized representation in the rule base for comparison and matching. Finally, a semantic similarity model, such as BERT or WordEmbeddings, is used to evaluate the similarity between the natural language rules and the existing rules.

[0052] Step S403: using a preset test set to test the accuracy of the detection rule or paradigm and provide feedback; wherein the preset test set contains verified detection rules or paradigms.

[0053] Specifically, first create a preset test set that includes various scenarios and situations. Then, write test cases and expected outputs that cover different types of detection rules or patterns. Finally, use the automated testing framework to execute the detection rules or patterns, and compare the execution results with the preset test set to evaluate the accuracy and effectiveness of the detection rules or patterns.

[0054] S50: Obtain a training set of the intelligent agent; the training set includes one or more of pictures, videos and sounds.

[0055] Specifically, training data is collected and prepared, including multi-modal data such as pictures, videos, and sounds, and labeled.

[0056] S60: Train the intelligent agent using the training set and the detection rules.

[0057] Specifically, we use the large-scale data sets collected above to train agents with various functions, and optimize the models to adapt to specific scenarios through transfer learning or adaptive learning. In addition, we also need to use historical data to train anomaly detection models so that they can identify situations that are inconsistent with known normal behaviors.

[0058] S70: Using the trained intelligent agent to manage the aircraft refueling operation area, and making corresponding management strategies according to the input information; the input information includes the user's natural language input or one or more of the pictures, videos and sounds of the aircraft refueling operation area.

[0059] It should be noted that the final output result of the agent is the information fusion of the detection results of each agent, and the risk situation is comprehensively judged through the decision algorithm, and corresponding preventive measures are taken. In this embodiment, the decision algorithm adopts a voting system or a weighted average method to obtain the final similarity score.

[0060] The management method for aircraft refueling operations provided in this embodiment ensures that the big model can correctly provide information through the professional knowledge base and knowledge graph in the field of aircraft refueling operations. In addition, by using natural language to define detection rules, the utilization rate of physical defense and human defense can be maximized. Finally, by using computer vision and big models to achieve multimodal recognition of risk scenarios, known risks can be effectively prevented, and legacy risks can be discovered through abnormal recognition, and management systems can be continuously improved.

[0061] Reference Figure 4 This embodiment also provides a management device for aircraft refueling operations, including: The first construction unit 100 is used to construct a knowledge graph about aircraft refueling operations. It should be noted that since the specific construction method and process have been explained in detail in step S10 of the above-mentioned aircraft refueling operation management method, they will not be repeated here.

[0062] The second construction unit 200 is used to construct a template library using the knowledge graph; similarly, since the specific construction method and process have been explained in detail in step S20 of the above-mentioned aircraft refueling operation management method, they will not be repeated here.

[0063] The third construction unit 300 is used to construct multiple intelligent entities with different functions based on the knowledge graph and the template library; similarly, since the specific construction method and process have been explained in detail in step S30 of the above-mentioned aircraft refueling operation management method, they will not be repeated here.

[0064] The fourth construction unit 400 is used to construct detection rules for determining risk scenarios based on the user's natural language input; similarly, since the specific construction method and process have been detailed in step S40 of the above-mentioned aircraft refueling operation management method, they will not be repeated here.

[0065] The acquisition unit 500 is used to acquire the training set of the intelligent agent; the training set includes one or more of pictures, videos and sounds; it should be noted that since the specific acquisition method and process have been detailed in step S50 of the above-mentioned aircraft refueling operation management method, they will not be repeated here.

[0066] The training unit 600 is used to train the intelligent agent using the training set and the detection rules. It should be noted that since the specific training method and process have been explained in detail in step S60 of the above-mentioned management method for aircraft refueling operations, they will not be repeated here.

[0067] The output unit 700 is used to manage the aircraft refueling operation area using the trained intelligent agent and make corresponding management strategies according to the input information; the input information includes the user's natural language input or one or more of the pictures, videos and sounds of the aircraft refueling operation area; it should be noted that since the specific output method and process have been explained in detail in step S70 of the above-mentioned aircraft refueling operation management method, they will not be repeated here.

[0068] In addition, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium may store a program, and when the program is executed, the program includes part or all of the steps of any one of the aircraft refueling operation management methods recorded in the above method embodiments.

[0069] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0070] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a memory, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0071] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable memory, and the memory may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0072] An exemplary flowchart for implementing management of aircraft refueling operations according to an embodiment of the present invention is described above with reference to the accompanying drawings. It should be noted that the large amount of details included in the above description are only exemplary illustrations of the present invention, rather than limitations of the present invention. In other embodiments of the present invention, the method may have more, fewer or different steps, and the order, inclusion, function, etc. of the steps may be different from those described and illustrated.

Claims

1. A method for managing aircraft refueling operations, characterized in that: include: Construct a knowledge graph about aircraft refueling operations; Using the knowledge graph to build a template library; Construct multiple intelligent agents with different functions according to the knowledge graph and the template library; Build detection rules for judging risk scenarios based on the user's natural language input; Obtaining a training set of the agent; the training set includes one or more of pictures, videos and sounds; Training the agent using the training set and the detection rule; The trained intelligent agent is used to manage the aircraft refueling operation area and make corresponding management strategies according to the input information; the input information includes the user's natural language input or one or more of the pictures, videos and sounds of the aircraft refueling operation area.

2. The management method according to claim 1, characterized in that: The construction of the knowledge graph about aircraft refueling operations includes: Determine the keyword list for knowledge terminal nodes; Establishing strongly associated knowledge point paragraph slices according to the keyword list; A knowledge graph across knowledge points is constructed based on the keyword list and strongly associated knowledge point segment slices.

3. The management method according to claim 2, characterized in that: The keyword list for determining the knowledge terminal node includes: Collect and maintain a keyword list covering all important concepts and terms in the field; using natural language processing tools to update and expand said keyword list; Determine the formal definition and context of the keywords to form a keyword list.

4. The management method according to claim 2, characterized in that: The step of establishing strongly associated knowledge point segment slices according to the keyword list includes: Analyze text corpora and extract combinations of knowledge points that frequently appear together; Using topic modeling techniques to identify and separate segments and establish correlations between combinations of knowledge points; Create paragraph indexes and mark strongly related knowledge points for quick retrieval and integration.

5. The management method according to claim 1, characterized in that: The step of constructing a template library using the knowledge graph includes: Collect common questions and answer patterns in the industry, and use the common questions and answer patterns to build a standardized template library; Use natural language generation technology to generate answers that conform to the template; The template library is combined with the knowledge graph, and the most suitable template is dynamically selected and filled according to the input question, so as to form the final template library.

6. The management method according to claim 1, characterized in that: The determination detection rule for judging risk scenarios based on the user's natural language input includes: Determine a detection rule or pattern based on the user's natural language input; Performing secondary confirmation of the detection rule or paradigm by similarity identification technology; The accuracy of the detection rule or paradigm is tested using a preset test set and feedback is provided; wherein the preset test set contains verified detection rules or paradigms.

7. The management method according to claim 6, characterized in that: The step of determining the detection rule or paradigm according to the user's natural language input includes: Build a rule language or paradigm that allows users to input natural language to define detection rules; Determining key prompt words and / or phrases, which serve as trigger conditions for establishing and modifying detection rules; Use natural language processing technology to parse the natural language rules provided by users.

8. The management method according to claim 6, characterized in that: The using of a preset test set to test the accuracy of the detection rule or paradigm and providing feedback includes: Build a pre-built test suite with a variety of scenarios and situations; Write test cases and expected outputs covering different types of detection rules or patterns; The detection rule or paradigm is executed using an automated testing framework, and the execution result is compared with the preset test set to evaluate the accuracy and effectiveness of the detection rule or paradigm.

9. A management device for aircraft refueling operations, characterized in that: include: The first construction unit is used to construct a knowledge graph about aircraft refueling operations; A second construction unit is used to construct a template library using the knowledge graph; A third construction unit is used to construct a plurality of intelligent agents with different functions according to the knowledge graph and the template library; A fourth construction unit, configured to construct a detection rule for determining a risk scenario according to a natural language input of a user; An acquisition unit, used to acquire a training set of the agent; the training set includes one or more of pictures, videos and sounds; A training unit, used for training the intelligent agent using the training set and the detection rule; The output unit is used to manage the aircraft refueling operation area using the trained intelligent agent and make corresponding management strategies according to the input information; the input information includes the user's natural language input or one or more of the pictures, videos and sounds of the aircraft refueling operation area.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for managing aircraft refueling operations as described in any one of claims 1 to 8 are implemented.

Images