Heterogeneous unmanned platform control instruction conversion method based on fields and knowledge graph
By building a control instruction conversion method based on field and knowledge graph, the problem of low control instruction conversion efficiency in heterogeneous unmanned clusters is solved, and efficient instruction conversion and system expansion is realized to ensure rapid response in complex environments.
Patent Information
- Application Number
- CN202411031799.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-30
- Publication Date
- 2024-11-12
AI Technical Summary
In heterogeneous unmanned clusters, the traditional control instruction conversion method has poor scalability, resulting in reduced conversion efficiency and cannot quickly adapt to the conversion requirements of multiple control instructions.
By constructing a control instruction transformation method based on field and knowledge graph, designing the control instruction ontology and knowledge graph, using the BERT-BiLSTM-CRF model to extract instruction knowledge, establishing a transformation knowledge graph of instruction structure and field mapping relationship, and realizing heterogeneous unmanned clusters Control the quick conversion between commands.
It significantly improves the efficiency of control instruction conversion and system scalability, reduces the cost of adapting to new instruction formats, simplifies system maintenance, and ensures rapid response and execution of control instructions in complex environments.
Smart Images

Figure CN118940829A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of heterogeneous unmanned cluster control, and specifically to a method for converting control instructions of a heterogeneous unmanned platform based on fields and knowledge graphs. Background Art
[0002] With the development of unmanned swarm technology, the problems that unmanned swarms need to solve are becoming more and more complicated. In many practical applications facing specific tasks, cross-domain unmanned systems with significant functional differences in sea, land, and air need to work together to complete the task. Compared with homogeneous systems with only a single type of unmanned system, cross-domain heterogeneous swarm intelligence systems can give full play to the advantages of different unmanned systems, form a stronger multi-dimensional information perception network and multi-directional reconnaissance capabilities, and be able to cope with more complex task requirements. However, there may be some differences in the control instructions for heterogeneous unmanned cluster nodes. When the command and control platform needs to control a heterogeneous unmanned cluster or heterogeneous unmanned nodes send control instructions to each other, how to quickly complete the conversion between control instructions is a problem that needs to be solved in the control of heterogeneous drone clusters.
[0003] In the execution of tasks by heterogeneous unmanned clusters, the control instructions of different types of unmanned nodes are different. In the process of quickly converting different control instructions, the traditional control instruction conversion method uses rule files or mapping tables to store the mapping relationship between different control instruction fields and intermediate formats. The rule files or mapping tables will increase with the addition of new control instructions, resulting in reduced conversion efficiency. Traditional control instruction conversion uses rules to complete specific control instruction conversion. The problem with this method is that it has poor scalability and is not suitable for the conversion of multiple control instructions. Summary of the invention
[0004] The purpose of the present invention is to provide a method for converting control instructions of heterogeneous unmanned platforms based on fields and knowledge graphs. By constructing the pattern layer and data layer of the graph, a conversion knowledge graph containing the mapping relationship between instruction structure and instruction fields is established; on this basis, a control conversion method based on the knowledge graph is proposed to achieve rapid conversion between heterogeneous unmanned cluster control instructions to solve the problems raised in the above background technology.
[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for converting control instructions of a heterogeneous unmanned platform based on fields and knowledge graphs, specifically comprising:
[0006] Step S1, determine the design scope of the instruction ontology, design the control instruction ontology, build the knowledge graph model layer, and provide a data framework for the knowledge graph;
[0007] Step S2, extracting entities from the data source, determining the relationship between entities according to the data framework, constructing a knowledge graph data layer, and forming a knowledge graph;
[0008] Step S3, constructing a control instruction conversion architecture based on the knowledge graph, and realizing rapid conversion between heterogeneous unmanned cluster control instructions according to the mapping relationship between fields in the knowledge graph.
[0009] Furthermore, the control instruction body includes five core classes: Type, Message, Fields, Value, and Payload Data;
[0010] Among them, Type indicates the type of heterogeneous unmanned cluster to which the control instruction belongs;
[0011] Message indicates the message type that the control instruction has;
[0012] Fields represents the mapping relationship between instruction instances;
[0013] Value represents the value of the field;
[0014] Payload Data indicates the payload information in the instruction.
[0015] Furthermore, the data source is the relevant knowledge of the control instructions, including semi-structured data and unstructured data, and the extraction process is specifically as follows:
[0016] Semi-structured data, using rules to extract knowledge;
[0017] For unstructured data, the BERT-BiLSTM-CRF model is used to extract knowledge.
[0018] Furthermore, the BERT-BiLSTM-CRF model extracts knowledge, specifically:
[0019] S21. The annotated corpus is passed through the BERT pre-trained language model, using the Transformer structure, and the weight coefficient matrix is adjusted according to the degree of association between words in the sentence to obtain the word representation and obtain the corresponding word vector. The formula is:
[0020]
[0021] Among them, Q, K, V are word vector matrices, d k The Transformer attention mechanism projects multiple linear transformations on Q, K, and V, and concatenates different attention results.
[0022] S22, input the word vector obtained in step S31 into the BiLSTM module, and process it using the bidirectional long short-term memory network LSTM. The specific output is:
[0023]
[0024] S23, using the CRF module to decode the output results of BiLSTM, obtain the predicted label sequence, and extract and classify each entity in the sequence, so as to complete the extraction of instruction knowledge, specifically:
[0025] The output score matrix of BiLSTM is the sequence X = (x1, x2, ..., x n ), p is a matrix of size n*k, P ij represents the score of the jth label of the ith word, for the predicted sequence Y=(y1,y2,…,y n ), the score function is:
[0026]
[0027] Among them, A represents the transfer score matrix, A ij Represents the fraction of label i being transferred to label j, and the size of A is k+2;
[0028] The probability of predicting sequence Y is:
[0029]
[0030] Taking the logarithm of both ends, the likelihood function of the predicted sequence is:
[0031]
[0032] in, represents the true label sequence, Y X represents all possible annotation sequences;
[0033] The output sequence with the maximum score after decoding is:
[0034]
[0035] Furthermore, the control instruction conversion architecture constructed based on the knowledge graph includes a knowledge layer and a conversion layer. The conversion layer is responsible for converting the received instruction information into another instruction through message parsing, instruction conversion and new control instruction generation steps; the knowledge layer is composed of an instruction conversion knowledge graph, which uses the instruction standard document as a data source, stores the field structure, field information and field mapping rule knowledge of each instruction, and provides knowledge and data support for the conversion layer.
[0036] Furthermore, step S3 is further:
[0037] S31, the instruction conversion program receives the original instruction data and parses it to obtain the original message information, and saves it in JSON format;
[0038] S32, mapping the original instruction field and the target instruction field in the JSON data to obtain the target message field and field value;
[0039] S33. According to the mapping relationship between fields in the instruction conversion knowledge graph, all fields in the target instruction message whose field attributes are fixed fields are queried in the instruction conversion knowledge graph. If a fixed field is missing, the fixed field of the target message is filled with default values to generate JSON data containing the target instruction information.
[0040] Furthermore, the mapping relationship between the original instruction field and the target instruction field includes: direct_mapping and indirect_mapping.
[0041] Beneficial effects of the invention: The core of the solution of the invention is to regard each field of the control instruction as an entity in the knowledge graph, and to reflect the mapping relationship between different control instruction fields through the relationship between entities; this solution reduces the tedious steps of converting the control instruction into an intermediate format in the traditional method, making the instruction conversion process more direct and efficient.
[0042] The mapping relationship between control instructions in this scheme is systematically stored in the knowledge graph, which not only makes the association between instructions more intuitive, but also provides great convenience for future expansion and updating. When the system needs to add new instruction conversion mapping relationships, it is only necessary to expand and update the existing knowledge graph accordingly, without large-scale modifications to the entire system, thereby significantly reducing the cost of adapting to the new instruction format.
[0043] This solution also improves the efficiency of conversion between control instructions, overcoming the need for complex adaptation work whenever a new instruction format is introduced in the traditional instruction conversion process. The knowledge graph-based method predefines and stores the mapping relationship between instructions, making the introduction of new instruction formats smoother and more efficient. This not only reduces the complexity of system maintenance, but also improves the overall command and control efficiency of the unmanned platform, ensuring rapid response and execution of various control instructions in complex and changing battlefield environments. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 It is a cross-domain interoperable instruction interaction knowledge graph.
[0045] Figure 2 It is a schematic diagram of the control instruction knowledge graph construction method.
[0046] Figure 3 This is the BERT-BiLSTM-CRF recognition framework diagram.
[0047] Figure 4 This is a control instruction conversion framework diagram based on the knowledge graph.
[0048] Figure 5 It is a schematic diagram of the control instruction conversion process.
[0049] Figure 6 It is a direct mapping relationship between fields.
[0050] Figure 7 It is an indirect mapping relationship between fields.
[0051] Figure 8 It is a protocol knowledge graph construction diagram.
[0052] Fig. 9 This is a control instruction diagram of a certain drone.
[0053] Fig.10 It is converted into an unmanned boat control command diagram. DETAILED DESCRIPTION
[0054] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is apparent to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features well known in the art are not described.
[0055] The present invention proposes a method for converting command and control instructions for heterogeneous unmanned platforms based on fields and knowledge graphs. It mainly includes a method for constructing a knowledge graph for converting heterogeneous control instructions and a method for converting heterogeneous control instructions based on a knowledge graph. This solution constructs a conversion knowledge graph containing a mapping relationship between instruction structure and instruction fields by constructing a graph model layer and a data layer. On this basis, a control conversion method based on a knowledge graph is proposed to achieve rapid conversion between heterogeneous unmanned cluster control instructions. The specific steps are:
[0056] Step S1, determine the design scope of the instruction ontology, design the control instruction ontology, build the knowledge graph model layer, and provide a data framework for the knowledge graph;
[0057] Step S2, extracting entities from the data source, determining the relationship between entities according to the data framework, constructing a knowledge graph data layer, and forming a knowledge graph;
[0058] Step S3, constructing a control instruction conversion architecture based on the knowledge graph, and realizing rapid conversion between heterogeneous unmanned cluster control instructions according to the mapping relationship between fields in the knowledge graph.
[0059] In a further embodiment, referring to Figure 1 ,Heterogeneous control instruction conversion is to achieve equivalent conversion of field information between two different control instructions under the premise of ensuring that the function remains unchanged. The instruction conversion knowledge graph takes fields as the main entities and displays the conversion mapping between control instructions through the relationship between fields. The logical structure of the knowledge graph consists of a schema layer and a data layer. The schema layer is the core of the knowledge graph and builds a framework for entities and their relationships. The data layer stores specific knowledge and expresses knowledge in the form of "entity-relationship-entity" triples and "entity-attribute-attribute value" pairs.
[0060] The knowledge required for control instruction conversion mostly comes from relevant design documents, which requires high professionalism and accuracy of relevant knowledge. Here, a top-down approach is used to construct the instruction knowledge graph. Figure 2 , the construction method of the control instruction conversion knowledge graph is specifically as follows:
[0061] Control instruction ontology design: Based on the concept of control instructions, five core classes are designed: Type, Message, Fields, Value, and Payload Data. Among them, Type indicates the type of heterogeneous unmanned cluster to which the control instruction belongs, such as uav1 (UAV 1), usv1 (Unmanned Boat 1), etc.; Message indicates the message type of the control instruction; Fields indicates the mapping relationship between instruction instances; Value indicates the value of the field; and Payload Data indicates the load information in the instruction.
[0062] The attributes of the ontology class can be divided into object attributes and data attributes. Object attributes are used to describe the relationship between classes. In this embodiment, the relationship between 8 core concepts of instructions is designed. The specific instruction ontology object attribute design is shown in the following table. Among them, there is a containment relationship between classes such as Type and Message, Type and Payload Data, that is, there is a mapping relationship between Fields and Fields, and Value and Value, which is divided into direct mapping and indirect mapping relationships between direct and in_direct.
[0063] kind relation kind Type has_message Message Type has_payload_data Payload Data Message has_field Fields Message message_mapping Message Fields has_value Value Fields direct_mapping Fields Fields indirect_mapping Fields Value value_mapping Value
[0064] Control instruction knowledge extraction: The relevant knowledge of control instructions mainly comes from the standard documents of instruction design, which mainly contain semi-structured data and unstructured data. For semi-structured data, relevant knowledge can be extracted directly using rules. For unstructured data, this embodiment proposes a BERT-BiLSTM-CRF model to extract relevant knowledge. The structure of the entire model is as follows: Figure 3As shown in the figure, the process is as follows: the annotated corpus is passed through the BERT pre-trained language model to obtain the corresponding word vector; the word vector is input into the BiLSTM module for further processing; the output result of the BiLSTM is decoded using the CRF module to obtain the predicted annotation sequence, and each entity in the sequence is extracted and classified to complete the extraction of instruction knowledge.
[0065] The BERT module is mainly used to extract stronger sentence semantic features for named entity recognition and obtain contextualized word vectors to improve entity recognition performance. Its key part is the deep network Transformer structure based on the "self-attention mechanism". The key part of the encoder is the self-attention mechanism, which mainly adjusts the weight coefficient matrix by the degree of association between words in the sentence to obtain the representation of the word. The formula is:
[0066]
[0067] Among them, Q, K, V are word vector matrices, d k is the Embedding dimension, and the attention mechanism projects Q, K, and V through multiple different linear transformations, and finally concatenates different Attention results.
[0068] Since the unidirectional LSTM model cannot process context information at the same time, in this embodiment, a new LSTM network, namely a bidirectional long short-term memory network, is used. The basic idea is to use forward and backward LSTM for each word sequence, and then merge the outputs at the same time. Therefore, for each moment, it corresponds to forward and backward information, and the output is shown in the following formula:
[0069]
[0070] In the named entity recognition task, BiLSTM is good at processing long-distance text information, but cannot process the dependency between adjacent tags. CRF can obtain an optimal prediction sequence through the relationship between adjacent tags, thus making up for the shortcomings of BiLSTM. In this embodiment, any sequence X = (x1, x2, ..., x n ) is the output score matrix of BiLSTM, p is a matrix of size n*k, where n is the number of words, k is the number of tags, and P ij represents the score of the jth label of the ith word, for the predicted sequence Y=(y1,y2,…,y n ), its score function is:
[0071]
[0072] Among them, A represents the transfer score matrix, A ijRepresents the fraction of label i being transferred to label j, and the size of A is k+2;
[0073] The probability of predicting sequence Y is:
[0074]
[0075] Taking the logarithm of both ends, the likelihood function of the predicted sequence is:
[0076]
[0077] In the formula, represents the true label sequence, Y X represents all possible annotation sequences;
[0078] The output sequence with the maximum score after decoding is:
[0079]
[0080] In a further embodiment, a control instruction conversion architecture based on a knowledge graph is provided, which is composed of a knowledge layer and a conversion layer. The conversion layer is responsible for converting the received instruction information into another instruction, including steps such as message parsing, instruction conversion, and generation of new control instructions. This process requires the knowledge layer to provide relevant knowledge to assist in instruction conversion. The knowledge layer is composed of an instruction conversion knowledge graph, which uses instruction standard documents as data sources, stores knowledge such as the structure of each instruction field, field information, and mapping rules between fields, and provides knowledge and data support for the conversion layer.
[0081] Reference Figure 4 After receiving the original message, the instruction conversion program obtains JSON data containing instruction information by parsing; maps the fields in the JSON data to obtain the corresponding target message fields and field values, and completes the default values of the fixed fields of the target message to generate JSON data containing the target instruction information; finally, generates new instruction information based on the JSON data.
[0082] Reference Figure 5 , the control instruction conversion process is as follows:
[0083] Parsing of original control instructions: The instruction conversion program receives and parses the original instruction, obtains the original message information such as the original instruction segment value, and saves it in the specified JSON format.
[0084] Mapping of original instructions and target instructions: In the control instruction conversion knowledge graph, if there is a "message_mapping" relationship between the type of the original instruction and the type of the target instruction, the conversion between the original instruction and the target instruction is established; there may be two mapping relationships between the original instruction field and the target instruction field: "direct_mapping" and "indirect_mapping".
[0085] direct_mapping indicates the direct mapping relationship between fields, such as Figure 6 As shown; indirect_mapping indicates the indirect mapping relationship between fields, indicating the relationship between their value_mapping value mappings, such as Figure 7 shown.
[0086] Generation of target instructions: Based on the mapping relationship between fields in the instruction conversion knowledge graph, the fields and field values in the target instruction that have a mapping relationship with the original instruction are obtained. However, in general, not all original instruction fields and target instruction fields have a mapping relationship. The target instruction message information obtained only through field mapping is incomplete. Therefore, it is necessary to query all fields in the target instruction message that have fixed field attributes in the instruction conversion knowledge graph to check whether all fixed fields are included in the target instruction information. If a fixed field is missing, its default value is used as the field value, and the corresponding key-value pair is supplemented, so that JSON data containing the target instruction information can be obtained.
[0087] In a further embodiment, referring to Figure 8 The knowledge graph constructed converts the command of a drone attacking a node into the control command of an unmanned boat attacking a node. The specific process is as follows:
[0088] The original control instruction is as follows Fig. 9 As shown, after parsing the original control instruction and obtaining the original message information such as the original instruction segment value, it is saved in JSON format;
[0089] Based on the control instruction conversion knowledge graph, the original instruction and the target instruction are mapped;
[0090] Check whether the target instruction information contains all fixed fields. If a fixed field is missing, use its default value as the field value and add the corresponding key-value pair to obtain JSON data containing the target instruction information, such as Fig.10 shown.
[0091] After the original control command is converted, the fields with different protocols such as longitude, latitude, and task id are converted accordingly. Among them, the superior id, since there is no corresponding field in the control command of the drone, the conversion result is empty, represented by "-". This embodiment completes the mutual control between different types of unmanned nodes by the command and control platform, and realizes the rapid conversion between heterogeneous unmanned cluster control commands.
[0092] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes may be made to it in form and detail without departing from the spirit and scope of the present invention as defined in the appended claims.
Claims
1. A control instruction conversion method for a heterogeneous unmanned platform based on fields and knowledge graphs, characterized in that: Specifically include: Step S1, determine the design scope of the instruction ontology, design the control instruction ontology, build the knowledge graph model layer, and provide a data framework for the knowledge graph; Step S2, extracting entities from the data source, determining the relationship between entities according to the data framework, constructing a knowledge graph data layer, and forming a knowledge graph; Step S3, constructing a control instruction conversion architecture based on the knowledge graph, and realizing rapid conversion between heterogeneous unmanned cluster control instructions according to the mapping relationship between fields in the knowledge graph.
2. The method for converting control instructions of a heterogeneous unmanned platform based on fields and knowledge graphs according to claim 1 is characterized in that: The control instruction body includes five core classes: Type, Message, Fields, Value, and PayloadData; Among them, Type indicates the type of heterogeneous unmanned cluster to which the control instruction belongs; Message indicates the message type that the control instruction has; Fields represents the mapping relationship between instruction instances; Value represents the value of the field; Payload Data indicates the payload information in the instruction.
3. The method for converting control instructions of a heterogeneous unmanned platform based on fields and knowledge graphs according to claim 1 is characterized in that: The data source is the relevant knowledge of control instructions, including semi-structured data and unstructured data. The extraction process is specifically as follows: Semi-structured data, using rules to extract knowledge; For unstructured data, the BERT-BiLSTM-CRF model is used to extract knowledge.
4. The method for converting control instructions of a heterogeneous unmanned platform based on fields and knowledge graphs according to claim 3 is characterized in that: The BERT-BiLSTM-CRF model extracts knowledge, specifically: S21. The annotated corpus is passed through the BERT pre-trained language model, using the Transformer structure, and the weight coefficient matrix is adjusted according to the degree of association between words in the sentence to obtain the word representation and obtain the corresponding word vector. The formula is: Among them, Q, K, V are word vector matrices, d k The Transformer attention mechanism projects multiple linear transformations on Q, K, and V, and concatenates different attention results. S22, input the word vector obtained in step S31 into the BiLSTM module, and process it using the bidirectional long short-term memory network LSTM. The specific output is: S23, using the CRF module to decode the output results of BiLSTM, obtain the predicted label sequence, and extract and classify each entity in the sequence, so as to complete the extraction of instruction knowledge, specifically: The output score matrix of BiLSTM is the sequence X = (x1, x2, ..., x n ), p is a matrix of size n*k, P ij represents the score of the jth label of the ith word, for the predicted sequence Y=(y1,y2,…,y n ), the score function is: Among them, A represents the transfer score matrix, A ij Represents the fraction of label i being transferred to label j, and the size of A is k+2; The probability of predicting sequence Y is: Taking the logarithm of both ends, the likelihood function of the predicted sequence is: in, represents the true label sequence, Y X represents all possible annotation sequences; The output sequence with the maximum score after decoding is:
5. The method for converting control instructions of a heterogeneous unmanned platform based on fields and knowledge graphs according to claim 1 is characterized in that: The control instruction conversion architecture constructed based on the knowledge graph includes a knowledge layer and a conversion layer. The conversion layer is responsible for converting the received instruction information into another instruction through message parsing, instruction conversion and new control instruction generation steps; the knowledge layer is composed of an instruction conversion knowledge graph, which uses the instruction standard document as a data source, stores the field structure, field information and field mapping rule knowledge of each instruction, and provides knowledge and data support for the conversion layer.
6. The method for converting control instructions of a heterogeneous unmanned platform based on fields and knowledge graphs according to claim 1 is characterized in that: Step S3 is further as follows: S31, the instruction conversion program receives the original instruction data and parses it to obtain the original message information, and saves it in JSON format; S32, mapping the original instruction field and the target instruction field in the JSON data to obtain the target message field and field value; S33. According to the mapping relationship between fields in the instruction conversion knowledge graph, all fields in the target instruction message whose field attributes are fixed fields are queried in the instruction conversion knowledge graph. If a fixed field is missing, the fixed field of the target message is filled with default values to generate JSON data containing the target instruction information.
7. The method for converting control instructions of a heterogeneous unmanned platform based on fields and knowledge graphs according to claim 6 is characterized in that: The mapping relationship between the original instruction field and the target instruction field includes: direct_mapping and indirect_mapping.
Citation Information
Cited By
Instruction conversion method of cross-chip platform
CN119292671A
