Phased Array Antenna Excitation Inversion Method Based on Design and Fault Knowledge Graphs

Abstract

The present invention discloses a phased array antenna excitation inversion method based on a design and fault knowledge graph, and its implementation steps are as follows: collect unstructured text data in the design and operation and maintenance stages of the phased array antenna; construct the ontology of the design and fault knowledge graph; generate a training set and a test set to obtain a qualified entity extraction model; input all text data into the qualified entity extraction model to obtain the knowledge triples of the design and fault knowledge graph; construct the design and fault knowledge graph, and fuse the array element excitation inversion of the design and fault knowledge graph; use the knowledge graph to invert the phased array antenna array element excitation. The present invention realizes the association of data in the design and operation and maintenance stages of the phased array antenna, realizes the inversion of information in the design and fault knowledge graph of the phased array antenna, and increases the reference of the knowledge graph for solving faults.

Description

Technical Field

[0001] The present invention belongs to the field of physical technologies, and further relates to a phased array antenna excitation inversion method based on a design and fault knowledge graph in the field of data processing technologies. The present invention can be used for phased array antenna excitation inversion that requires fault and design information. Background Art

[0002] In the process from the design to the operation and maintenance of phased array antennas, various design information and fault information are scattered and weakly presented, and are basically recorded in text form. Relevant personnel can only understand the content of design and operation and maintenance by searching for various data materials, and there is no direct connection between texts. The analysis of phased array antenna fault information involves the participation of operation and maintenance personnel and design personnel, and is a complex dynamic process. In order to ensure the normal operation of the equipment system and improve the efficiency of fault analysis, using a knowledge graph for the analysis of phased array antenna faults can increase the relevance between faults and design information. Designers can more efficiently query fault information, design information, and recommended solutions through the knowledge graph. In existing fault analysis methods based on knowledge graphs, due to the lack of information association or imperfect recommended solutions, a design and fault knowledge graph with an inversion function and associated design information and fault information cannot be obtained.

[0003] Sun Yat-sen University proposed a fault analysis method in its patent document "A Fault Analysis Method for Cloud Native Systems Based on a Knowledge Graph" (application number 202011554734.6, publication number CN 112540832 A). The implementation steps of this method include: first, obtaining original data and constructing a knowledge graph based on the original data to obtain graph data; second, performing anomaly detection on the graph data through an anomaly detection model to obtain anomaly nodes; third, calculating the similarity between the anomaly nodes and the replica nodes corresponding to the anomaly nodes, and performing fault root cause location based on the similarity. The disadvantage of this method is that ontology construction is not carried out before constructing the knowledge graph, and it is only limited to constructing a general knowledge graph with a wide knowledge coverage. The knowledge and data generated during the design, operation, and maintenance of phased array antennas are usually knowledge in a specific professional field, and unstructured data usually lacks associations and requires ontology construction first. This method does not specify entity types and relationship types for entity extraction and relationship extraction, and it is difficult to apply to the field of phased array antennas.

[0004] Guangzhou Huitong Guoxin Technology Co., Ltd. discloses a fault handling recommendation method in its patent document "A Method and Device for Recommending Power Grid Fault Handling Plans Based on Knowledge Graph" (application number 202210328887.1, publication number CN 114756686 A). The implementation steps of this method are as follows: The first step is information acquisition; the second step is to organize the fault information, convert its format, and extract its fault keywords; the third step is to extract the fault handling plans in the database according to the fault keywords; the fourth step is to match the extracted fault handling plans, and if the match fails, a prompt will be given. The deficiency of this fault handling recommendation method is that the recommended fault handling solutions in the knowledge graph can only use past fault handling solutions as a reference, and cannot perform inversion based on information to obtain the element excitation for specifically solving the fault, and it is not applicable to the phased array antenna field that requires specific element excitation according to different design information. Summary of the Invention

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned existing technologies and propose a phased array antenna excitation inversion method based on design and fault knowledge graphs, which is used to solve the lack of association between unstructured data of design information and unstructured data of fault information, as well as the problem that the recommended solution can only use past fault solutions as a reference and cannot perform inversion based on specific design information and fault information to obtain the element excitation for specifically solving the fault.

[0006] The idea of realizing the purpose of the present invention is to collect unstructured text data in the operation and maintenance and design stages of phased array antennas, construct the ontology of unstructured text data, which makes the relationship between entities more specific and can directly form associations between the extracted entities, generate a training set and a test set for training and obtain an entity extraction model. The trained entity extraction model extracts entities from unstructured text data, constructs a design and fault knowledge graph, fuses the design and fault knowledge graph for element excitation inversion, performs inversion on the phased array antenna element excitation according to the fault information using the knowledge graph, and recommends the inverted element excitation to the design information to solve the fault.

[0007] To achieve the above purpose, the steps of the technical solution of the present invention are as follows:

[0008] Step 1: Collect unstructured text data generated in the design and operation and maintenance stages of phased array antennas, where the unstructured text data includes design documents, diagnostic, and maintenance records;

[0009] Step 2: Construct the ontology of the design and fault knowledge graph:

[0010] According to the unstructured text data generated during the design and operation and maintenance phases of the phased array antenna, 17 entity types and 4 relationship types are respectively set, and the 17 entity types and 4 relationship types are constructed into 16 "head entity - relationship - tail entity" knowledge triples. All entity types, relationship types, and knowledge triples form the ontology of the design and fault knowledge graph;

[0011] Step 3, generate the training set and the test set:

[0012] Step 3.1, randomly select 20% of the text data from the unstructured text data, and label the entity type tags of each selected text data;

[0013] Step 3.2, generate the training set and the test set according to the ratio of 7:3 from the labeled text data;

[0014] Step 4, obtain a qualified entity extraction model:

[0015] Step 4.1, input the training set into the entity extraction model based on the bidirectional long short - term memory network, and use the stochastic gradient descent method to iteratively update the parameters in the entity extraction model until the loss function converges, obtaining a trained entity extraction model;

[0016] Step 4.2, input the test set into the trained entity extraction model, output the predicted entity type tags of each text data in the test set, calculate the accuracy rate of the entity type tags of all predicted text data output by the entity extraction model. When the extraction accuracy rate of this extraction model reaches 80%, obtain a qualified entity extraction model;

[0017] Step 5, construct the design and fault knowledge graph:

[0018] Step 5.1, input all the text data in the unstructured text data into the qualified entity extraction model, and output the entities and their corresponding entity types of each text data;

[0019] Step 5.2, associate the entities in each text data according to the entity types according to the knowledge triples in the ontology, obtaining the knowledge triples of the design and fault knowledge graph of this text data;

[0020] Step 5.3, store all the knowledge triples into the Neo4j graph database. Through the generation instruction of the Neo4j graph database, take the entities of each knowledge triple as a corresponding node, and the relationship as an edge connecting the head entity node and the tail entity node, obtaining the design and fault knowledge graph;

[0021] Step 6, design and fault knowledge graph fusion array element excitation inversion:

[0022] Step 6.1, extract all entities with entity types of: number of array elements, array element coordinates, array element excitation, channel fault description, and allowable error of sidelobe level from the fault and design knowledge graph. Combine the channel fault description and the array element excitation to form a failed array element excitation. Combine a set of related array element excitations, channel fault descriptions, allowable error of sidelobe level, and failed array element excitations to form a set of input parameters for array element excitation inversion.

[0023] Step 6.2, select a set of input parameters for array element excitation inversion that have not been selected before. Substitute the array element excitation in the selected input parameters for array element excitation inversion into the direction pattern function of the antenna array to obtain the data of the radiation direction pattern. Extract one sidelobe level on each of the left and right sides of the main lobe in the azimuth plane and elevation plane from this data. Combine all the extracted data to form the pre-failure direction pattern parameters.

[0024] Step 6.3, using the sidelobe level and main lobe beam pointing in the pre-failure direction pattern parameters as the targets, substitute the failed array element excitation in the input parameters for array element excitation inversion of the same group into the evolutionary algorithm, and iteratively update the array element excitation of the current iteration until the iteration termination condition is met. Take the output array element excitation of the current iteration as the inverted array element excitation.

[0025] Step 6.4, associate the inverted array element excitation with the channel fault description in the input parameters for array element excitation inversion of the same group. Take each associated inverted array element excitation and the "channel fault description" in the channel fault description as the head entity in the "head entity - relationship - tail entity" knowledge triple, and the "inverted array element excitation" as the tail entity to obtain a "channel fault description - inference - inverted array element excitation" triple. Import this triple into the Neo4j graph database for storage.

[0026] Step 6.5, determine whether all input parameters for array element excitation inversion have been selected. If so, execute Step 7; otherwise, execute Step 6.2.

[0027] Step 7, use the knowledge graph to invert the array element excitation of the phased array antenna:

[0028] Step 7.1, collect the text data when the phased array antenna fails during the operation and maintenance phase. Use the same methods as in Step 5.1 and Step 5.2 to obtain the entities of the fault text data. According to the entity type of each entity, find the "relationship" in the knowledge triple corresponding to this entity type in the ontology, and associate the entities of the fault data to obtain the knowledge triple of this fault text data. Update the fault information in the design and fault knowledge graph with the knowledge triple to obtain the design and fault knowledge graph with updated fault information.

[0029] Step 7.2: According to the "phased array antenna name" entity in the fault data, extract the entities corresponding to the "phased array antenna name" entity in the updated design and fault knowledge graph, including element excitation, channel fault description, allowable error of sidelobe level, and failed element excitation; form a set of input parameters for element excitation inversion by combining the associated element excitation, channel fault description, allowable error of sidelobe level, and failed element excitation.

[0030] Step 7.3: Use the same method as in Steps 6.2 and 6.3 to obtain the inverted element excitation, and associate the element excitation with the extracted channel fault description; use the "channel fault description" in the associated inverted element excitation and channel fault description as the head entity, and the "inverted element excitation" as the tail entity in the "head entity - relationship - tail entity" knowledge triple, to obtain the "channel fault description - reasoning - inverted element excitation" triple, and import this triple into the Neo4j graph database to update the relevant data, and obtain the design and fault knowledge graph after updating the inverted element excitation.

[0031] Step 7.4: The designer searches for the inverted element excitation in the design and fault knowledge graph after updating the inverted element excitation, and modifies the phased array antenna element excitation according to the inverted element excitation to obtain the array design information of the phased array antenna after the fault.

[0032] The present invention has the following advantages compared with the prior art:

[0033] First, the present invention constructs an ontology of the design and fault knowledge graph for the design and operation and maintenance stages of phased array antennas, overcoming the deficiency of the prior art that when extracting unstructured text data, due to the lack of association between the unstructured design information and unstructured fault information in the design and operation and maintenance stages of phased array antennas, it is impossible to effectively extract the relationships, resulting in unassociated data. The present invention can completely obtain valuable information in the unstructured data, realize the association of a large amount of data in the design and operation and maintenance stages of phased array antennas, enable the operation and maintenance personnel and designers to query both the design information in the design stage and the fault information in the operation and maintenance stage, and improve the efficiency of solving phased array antenna faults.

[0034] Second, the present invention inverses the information in the design and fault knowledge graph for the design and operation and maintenance stages of phased array antennas, and adds the inverted element excitation to the knowledge graph, overcoming the deficiency of the prior art that the recommended solutions for faults can only use past fault measures as references. The present invention can inverse according to different information to obtain the specific element excitation for solving faults, increasing the reference value of the knowledge graph for designers to solve faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is a flowchart of the present invention;

[0036] Figure 2 This is the flow chart of the fusion of design and fault knowledge graph for array element excitation inversion in the present invention. Specific implementation manners

[0037] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0038] Refer to Figure 1 for a further description of the implementation steps of the embodiments of the present invention.

[0039] Step 1: Collect unstructured text data generated during the design and operation and maintenance phases of the phased array antenna. This unstructured text data includes design documents, diagnostic records, and maintenance records.

[0040] Take 500 pieces of unstructured text data generated during the design and operation and maintenance phases of the phased array antenna of a certain company as the source.

[0041] Step 2: Construct the ontology of the design and fault knowledge graph according to the unstructured text data generated during the design and operation and maintenance phases of the phased array antenna.

[0042] Set entity types, relationship types, and "head entity - relationship - tail entity" knowledge triples.

[0043] The ontology of the text data constructed in the embodiments of the present invention includes 16 entity types and 4 relationship types.

[0044] The 17 entity types are: phased array antenna name, radio frequency network, power distribution, power module, array unit, T / R component, SMA connector, array design information, number of array elements, array element coordinates, array element excitation, allowable error of sidelobe level, operating wavelength, channel fault information, channel fault description, failure rate, and failure cause.

[0045] The 4 relationship types are: construct, information, contain, and inference.

[0046] Use 17 entity types and 4 relationship types to construct 16 "head entity - relationship - tail entity" knowledge triples respectively as shown in Table 1. Each row in Table 1 represents a knowledge triple.

[0047] The construction of 16 "head entity - relation - tail entity" knowledge triples from 17 entity types and 4 relation types in the ontology refers to: constructing 16 "head entity - relation - tail entity" knowledge triples including "Radio Frequency Network - Phased Array Antenna Name - Construction", "Power Distribution - Phased Array Antenna Name - Construction", "Power Module - Phased Array Antenna Name - Construction", "Array Unit - Phased Array Antenna Name - Construction", "T / R Component - Phased Array Antenna Name - Construction", "SMA Connector - Phased Array Antenna Name - Construction", "Array Design Information - Phased Array Antenna Name - Information", "Array Design Information - Number of Array Elements - Inclusion", "Array Design Information - Array Element Coordinates - Inclusion", "Array Design Information - Array Element Excitation - Inclusion", "Array Design Information - Allowable Error of Sidelobe Level - Inclusion", "Array Design Information - Operating Wavelength - Inclusion", "Channel Fault Information - Inclusion - Channel Fault Description", "Channel Fault Description - Inclusion - Failure Probability", "Channel Fault Description - Inclusion - Fault Inducement", "Fault Inducement - Inference - Module Name" from 17 entity types including "Phased Array Antenna Name", "Radio Frequency Network", "Power Distribution", "Power Module", "Array Unit", "T / R Component", "Array Design Information", "Number of Array Elements", "Array Element Coordinates", "Array Element Excitation", "Allowable Error of Sidelobe Level", "Operating Wavelength", "Channel Fault Information", "Channel Fault Description", "Failure Rate", and "Fault Inducement" and 4 relation types including "Construction", "Information", "Inclusion", and "Inference".

[0048] Table 1 List of Ontology Knowledge Triples

[0049]

[0050] Step 3: Generate the training set and the test set.

[0051] Randomly select 100 pieces of text data from the unstructured text data, and use the BMEO annotation method to perform entity annotation on the training set and the test set according to the entity types in the ontology, label the entity type tags of the text data to obtain the labeled text data. For example, if there is a word of the type "Channel Fault Description" in a certain text data, then "B - Channel Fault Description" represents the first character of the word, "M - Channel Fault Description" represents all the intermediate characters except the first and the last of the word, "E - Fault Mode" represents the last character of the word, and the label "O" represents that the character is not in the entity. Compose 70 pieces of the labeled text data into the training set, and the remaining 30 pieces of the text data into the test set.

[0052] Step 4: Obtain a qualified entity extraction model.

[0053] Step 4.1: Input the training set into the entity extraction model based on the bidirectional long short-term memory network. Use the stochastic gradient descent method to iteratively update the parameters in the model until the loss function converges, and obtain the trained entity extraction model.

[0054] The network loss function is as follows:

[0055]

[0056] Among them, MSE represents the loss value function between the label predicted by the model and the label annotated by the entity in the text data. n represents the number of text data. y i represents the label annotated by the i-th entity in the training set, represents the i-th label predicted by the output of the training set model.

[0057] The entity extraction model based on the bidirectional long short-term memory network is a model constructed by existing technologies. For its structure and the setting of initial parameters, refer to the description in the paper "Knowledge Graph Construction Technology and Application for Aircraft Power System Fault Diagnosis" published by Nie Tongpan et al. (Acta Aeronautica et Astronautica Sinica, 2021, 42).

[0058] Step 4.2: Input the test set into the trained entity extraction model, output the entity type labels predicted for each text data in the test set, calculate the accuracy of the entity type labels of all predicted text data output by the entity extraction model. When the extraction accuracy of the extraction model reaches 80%, obtain a qualified entity extraction model.

[0059] The accuracy calculation is as follows:

[0060]

[0061] Among them, P is the accuracy of the entity extraction model label, T is the number of labels where the label output by the entity extraction model is consistent with the test set annotation label, and F is the number of labels where the label output by the entity extraction model is inconsistent with the test set annotation label.

[0062] Step 5: Construct the design and fault knowledge graph.

[0063] Step 5.1: Input all text data in the unstructured text data into the qualified entity extraction model, and output the entities and their corresponding entity types of each text data.

[0064] Step 5.2: Associate the entities in each text data according to the entity types in the knowledge triples of the ontology to obtain the knowledge triples of the design and fault knowledge graph of this text data.

[0065] For example, the entities extracted from the text data "The failure of Channel 1 was caused by the failure of the SMA connector." are "Channel 1 failure" and "SMA connector". Among them, the entity type of "Channel 1 failure" is channel failure description, and the entity type of "SMA connector" is module. According to the triple of channel failure description - failure cause - failure location in the ontology, the triple of Channel 1 failure - failure cause - SMA connector is constructed.

[0066] Step 5.3, store all the knowledge triples in the Neo4j graph database. Through the generation instruction of the Neo4j graph database, take the entities of each knowledge triple as corresponding nodes, and the relationship as an edge connecting the head entity node and the tail entity node to obtain the design and fault knowledge graph.

[0067] The Neo4j is a high-performance open-source non-relational graph database developed by Neo4j, Inc.

[0068] Step 6, the design and fault knowledge graph fuses the array element excitation inversion.

[0069] Step 6.1, extract all entities with entity types of: number of array elements, array element coordinates, array element excitation, channel failure description, and allowable error of sidelobe level from the fault and design knowledge graph. Combine the channel failure description with the array element excitation to form the failed array element excitation; form a set of input parameters for array element excitation inversion by combining a group of related array element excitations, channel failure descriptions, allowable error of sidelobe level, and failed array element excitations.

[0070] Step 6.2, select a set of input parameters for array element excitation inversion that have not been selected before; substitute the array element excitation into the radiation pattern function of the antenna array to obtain the data of the radiation pattern, extract the positions of the first zeros on the left and right of the main lobe in the azimuth plane and a maximum sidelobe level from this data, and at the same time extract the positions of the first zeros on the left and right of the main lobe in the elevation plane and a maximum sidelobe level. Combine all the extracted data to form the parameters of the radiation pattern before failure.

[0071] The radiation pattern function of the antenna array is as follows:

[0072]

[0073] where is the array element field strength direction function, θ and are the elevation angle and azimuth angle respectively, N is the number of array elements, I n is the amplitude of the array element excitation of the nth array element, e is the natural constant, j is the imaginary unit, π is the pi, λ is the operating wavelength, x n represents the x coordinate of the nth array element coordinate, y n represents the y coordinate of the nth array element coordinate, b nis the phase of the element excitation of the nth element.

[0074] Step 6.3, set the population size NP = 200, the number of iterations G = 200, the crossover probability Pc = 0.8, and the mutation probability Pm = 0.05 of the genetic algorithm. The chromosome length in the population is determined by the number of elements. Set the failed element excitation obtained in Step 7.1 as the first chromosome in the initial population, and randomly generate other chromosomes.

[0075] Step 6.4, successively perform roulette wheel selection, probability crossover with a probability value of Pc, and probability mutation with a probability value of Pm to generate a new population for the current iteration. Substitute the chromosome into the radiation pattern function of the antenna array to obtain the data of the radiation pattern. Extract the positions of the first null on the left and right sides of the main lobe in the azimuth plane and a maximum sidelobe level from this data. At the same time, extract the positions of the first null on the left and right sides of the main lobe in the elevation plane and a maximum sidelobe level. Combine all the extracted data into the parameters of the radiation pattern corresponding to the chromosome, and calculate the fitness of the chromosome. Always retain the chromosome with the lowest fitness in the population as the first chromosome in the population.

[0076] The fitness function is as follows:

[0077] fitness = 1 / (|FSLL_AZ - FSLL_AZ_set| + |FSLL_EL - FSLL_EL_set| + 0.9|NULL_AZ_1 - NULL_AZ_set1| + 0.9|NULL_AZ_2 - NULL_AZ_set2| + 0.9|NULL_EL_2 - NULL_EL_set2| + 0.9|NULL_EL_2 - NULL_EL_set2|)

[0078] Among them, fitness is the fitness of the chromosome, FSLL_AZ is the maximum sidelobe level in the azimuth plane among the pattern parameters corresponding to the chromosome, FSLL_AZ_set is the maximum sidelobe level in the azimuth plane among the pattern parameters before failure, FSLL_EL is the maximum sidelobe level in the elevation plane among the pattern parameters corresponding to the chromosome, FSLL_EL_set is the maximum sidelobe level in the elevation plane among the pattern parameters before failure, NULL_AZ_1 is the position of the first zero on the left side of the main lobe in the azimuth plane among the pattern parameters corresponding to the chromosome, NULL_AZ_set1 is the position of the first zero on the left side of the main lobe in the azimuth plane among the pattern parameters before failure, NULL_AZ_2 is the position of the first zero on the right side of the main lobe in the azimuth plane among the pattern parameters corresponding to the chromosome, NULL_AZ_set2 is the position of the first zero on the right side of the main lobe in the azimuth plane among the pattern parameters before failure, NULL_EL_1 is the position of the first zero on the left side of the main lobe in the elevation plane among the pattern parameters corresponding to the chromosome, NULL_EL_set1 is the position of the first zero on the left side of the main lobe in the elevation plane among the pattern parameters before failure, NULL_EL_2 is the position of the first zero on the right side of the main lobe in the elevation plane among the pattern parameters corresponding to the chromosome, and NULL_EL_set2 is the position of the first zero on the right side of the main lobe in the elevation plane among the pattern parameters before failure.

[0079] Repeat step 6.4 until the iteration termination condition is met. Extract the first chromosome from the population at the end of the iteration as the inverted element excitation.

[0080] The iteration termination condition is a situation that simultaneously satisfies the following two conditions.

[0081] Condition 1: The error between the sidelobe level in the elevation plane among the pattern parameters corresponding to the chromosome with the lowest fitness and the sidelobe level in the elevation plane among the pattern parameters before failure is within the allowable error of the sidelobe level.

[0082] Condition 2: The error between the sidelobe level in the azimuth plane among the pattern parameters corresponding to the chromosome with the lowest fitness and the sidelobe level in the azimuth plane among the pattern parameters before failure is within the allowable error of the sidelobe level.

[0083] Step 6.5, associate the inverted element excitation with the channel fault description in the input parameters inverted from the element excitations of the same group; take each associated inverted element excitation and the "channel fault description" in the fault description as the head entity in the "head entity - relationship - tail entity" knowledge triple, and the "inverted element excitation" as the tail entity to obtain a "fault description - reasoning - inverted element excitation" triple, and import this triple into the Neo4j graph database for storage. The process is as Figure 2 shown.

[0084] Step 6.6, determine whether all the input parameters for the inversion of the array element excitations have been selected. If so, execute Step 7; otherwise, execute Step 6.2. 500 inverted array element excitations are obtained for all the input parameters for the inversion of the array element excitations.

[0085] Step 7, use the knowledge graph to invert the array element excitations of the phased array antenna.

[0086] Step 7.1, collect the text data when the phased array antenna fails during the operation and maintenance phase. Using the same methods as in Steps 5.1 and 5.2, obtain the entities of the fault text data. According to the entity type of each entity, find the "relationship" in the knowledge triples corresponding to the entity type in the ontology, associate the entities of the fault data, obtain the knowledge triples of the fault text data, and update the fault information in the design and fault knowledge graphs with the knowledge triples to obtain the design and fault knowledge graphs with updated fault information.

[0087] Step 7.2, according to the entity of "phased array antenna name" in the fault data, extract the entities in the updated design and fault knowledge graphs corresponding to the entity type of the "phased array antenna name" entity: array element excitation, channel fault description, and allowable error of sidelobe level. Combine the associated array element excitation, channel fault description, allowable error of sidelobe level, and failed array element excitation into a set of input parameters for the inversion of the array element excitations.

[0088] Step 7.3, use the same methods as in Steps 6.2 and 6.3 to obtain the inverted array element excitations, and associate the array element excitations with the extracted channel fault descriptions. Use the "channel fault description" in the associated inverted array element excitation and channel fault description as the head entity, and the "inverted array element excitation" as the tail entity in the "head entity - relationship - tail entity" knowledge triple to obtain the "channel fault description - inference - inverted array element excitation" triple, and import this triple into the Neo4j graph database to update the relevant data to obtain the design and fault knowledge graphs with updated inverted array element excitations.

[0089] Step 7.4, the designer searches for the inverted array element excitations in the design and fault knowledge graphs with updated inverted array element excitations, and modifies the array element excitations of the phased array antenna according to the inverted array element excitations to obtain the array design information of the phased array antenna after the fault.

Claims

1. A phased array antenna excitation inversion method based on a design and fault knowledge graph, characterized in that Construct the ontology of the design and fault knowledge graph, fuse the design and fault knowledge graph for array element excitation inversion, and use the knowledge graph to perform inversion on the array element excitation of the phased array antenna; the steps of the inversion method are as follows: Step 1, collect the unstructured text data generated during the design and operation and maintenance phases of the phased array antenna. The unstructured text data includes design documents and diagnostic and maintenance records; Step 2, construct the ontology of the design and fault knowledge graph: According to the unstructured text data generated during the design and operation and maintenance phases of the phased array antenna, set 17 entity types and 4 relationship types respectively, construct 16 "head entity-relationship-tail entity" knowledge triples from the 17 entity types and 4 relationship types, and form the ontology of the design and fault knowledge graph with all entity types, relationship types, and knowledge triples; Step 3, generate the training set and test set: Step 3.1, randomly select 20% of the text data from the unstructured text data, and label the entity type labels of each selected text data; Step 3.2, generate the training set and test set according to the ratio of 7:3 for the labeled text data; Step 4, obtain a qualified entity extraction model: Step 4.1, input the training set into the entity extraction model based on the bidirectional long short-term memory network, use the stochastic gradient descent method to iteratively update the parameters in the entity extraction model until the loss function converges, and obtain the trained entity extraction model; Step 4.2, input the test set into the trained entity extraction model, output the predicted entity type labels of each text data in the test set, calculate the accuracy rate of the entity type labels of all predicted text data output by the entity extraction model, and when the extraction accuracy rate of the extraction model reaches 80%, obtain a qualified entity extraction model; Step 5, construct the design and fault knowledge graph: Step 5.1, input all the text data in the unstructured text data into the qualified entity extraction model, and output the entities of each text data and their corresponding entity types; Step 5.2, associate the entities in each text data according to the entity types according to the knowledge triples in the ontology to obtain the knowledge triples of the design and fault knowledge graph of the text data; Step 5.3, store all the knowledge triples in the Neo4j graph database. Through the generation instructions of the Neo4j graph database, use the entities of each knowledge triple as a corresponding node, and the relationship as an edge connecting the head entity node and the tail entity node to obtain the design and fault knowledge graph; Step 6, fuse the design and fault knowledge graph for array element excitation inversion: Step 6.1, extract all entities with entity types of: number of array elements, array element coordinates, array element excitation, channel fault description, and allowable error of sidelobe level from the fault and design knowledge graph. Combine the channel fault description and the array element excitation to form the failed array element excitation; form a set of input parameters for array element excitation inversion with a group of related array element excitations, channel fault descriptions, allowable error of sidelobe level, and failed array element excitations; Step 6.2, select a set of input parameters for array element excitation inversion that has not been selected before; Substitute the element excitations in the input parameters of the selected element excitation inversion into the radiation pattern function of the antenna array to obtain the data of the radiation pattern. Extract one sidelobe level on each of the left and right sides of the main lobe in the azimuth plane and elevation plane from this data, and form the pre-failure pattern parameters with all the extracted data; Step 6.3: Take the sidelobe level and main lobe beam pointing in the pre-failure pattern parameters as the targets, substitute the failed element excitations in the input parameters of the same set of element excitation inversion into the evolutionary algorithm, and iteratively update the element excitations of the current iteration until the iteration termination condition is met. Take the output element excitations of the current iteration as the inverted element excitations; Step 6.4: Associate the inverted element excitations with the channel fault description in the input parameters of the same set of element excitation inversion; Take each associated inverted element excitation and the "channel fault description" in the channel fault description as the head entity in the "head entity - relation - tail entity" knowledge triple, and the "inverted element excitation" as the tail entity to obtain a "channel fault description - inference - inverted element excitation" triple, and import this triple into the Neo4j graph database for storage; Step 6.5: Determine whether all the input parameters of the element excitation inversion have been selected. If so, execute Step 7; otherwise, execute Step 6.2; Step 7: Invert the element excitations of the phased array antenna using the knowledge graph: Step 7.1: Collect the text data when the phased array antenna fails during the operation and maintenance stage. Use the same methods as in Steps 5.1 and 5.2 to obtain the entities of the fault text data. According to the entity type of each entity, find the "relation" in the knowledge triple corresponding to this entity type in the ontology, and associate the entities of the fault data to obtain the knowledge triple of this fault text data. Update the fault information in the design and fault knowledge graph with the knowledge triple to obtain the design and fault knowledge graph with updated fault information; Step 7.2: According to the "phased array antenna name" entity of the fault data, extract the entities in the updated design and fault knowledge graph corresponding to this "phased array antenna name" entity with the entity types of: element excitation, channel fault description, sidelobe level allowable error, and failed element excitation; Form a set of input parameters for the element excitation inversion with the associated element excitations, channel fault descriptions, sidelobe level allowable errors, and failed element excitations; Step 7.3: Use the same methods as in Steps 6.2 and 6.3 to obtain the inverted element excitations, and associate the element excitations with the extracted channel fault descriptions; Take the associated "channel fault description" and the "inverted element excitation" in the channel fault description as the head entity in the "head entity - relation - tail entity" knowledge triple, and the "inverted element excitation" as the tail entity to obtain this "channel fault description - inference - inverted element excitation" triple, and import this triple into the Neo4j graph database to update the relevant data to obtain the design and fault knowledge graph with updated inverted element excitations; Step 7.4, the designer searches for the inverted element excitation in the updated design and fault knowledge graph after the inversion, and modifies the phased array antenna element excitation according to the inverted element excitation to obtain the array design information of the phased array antenna after the fault.

2. The phased array antenna excitation inversion method based on the design and fault knowledge graph according to claim 1, wherein, The construction of the 17 entity types and 4 relationship types in the ontology into 16 "head entity - relationship - tail entity" knowledge triples described in Step 2 means: constructing the 17 entity types including "phased array antenna name", "radio frequency network", "power distribution", "power module", "array unit", "T / R component", "array design information", "number of array elements", "element coordinates", "element excitation", "allowable error of sidelobe level", "operating wavelength", "channel fault information", "channel fault description", "failure rate", and "fault cause" and the 4 relationship types including "construction", "information", "contains", and "inference" into 16 "head entity - relationship - tail entity" knowledge triples: "radio frequency network - phased array antenna name - construction", "power distribution - phased array antenna name - construction", "power module - phased array antenna name - construction", "array unit - phased array antenna name - construction", "T / R component - phased array antenna name - construction", "SMA connector - phased array antenna name - construction", "array design information - phased array antenna name - information", "array design information - number of array elements - contains", "array design information - element coordinates - contains", "array design information - element excitation - contains", "array design information - allowable error of sidelobe level - contains", "array design information - operating wavelength - contains", "channel fault information - contains - channel fault description", "channel fault description - contains - failure probability", "channel fault description - contains - fault cause", "fault cause - inference - module name".

3. The phased array antenna excitation inversion method based on a design and fault knowledge graph according to claim 1, characterized in that The loss function described in Step 4.1 is as follows: Among them, MSE represents the loss value function between the label predicted by the model and the label annotated by the text data entity, n represents the number of text data, and y i represents the label annotated by the i-th entity in the training set, represents the i-th label predicted by the model output of the training set.

4. The phased array antenna excitation inversion method based on a design and fault knowledge graph according to claim 1, wherein The accuracy calculation described in Step 4.2 is as follows: Where P is the accuracy of the entity extraction model label, T is the number of labels where the labels output by the entity extraction model are consistent with the labels in the test set annotation, and F is the number of labels where the labels output by the entity extraction model are inconsistent with the labels in the test set annotation.

5. The phased array antenna excitation inversion method based on a design and fault knowledge graph according to claim 1, characterized in that The iteration termination condition described in Step 7.3 is the situation where the following two conditions are simultaneously met: Condition 1, the error between the azimuth plane sidelobe level in the pattern parameters corresponding to the inverted element excitation and the azimuth plane sidelobe level in the pattern parameters before failure is within the allowable error of the sidelobe level; Condition 2, the error between the elevation plane sidelobe level in the pattern parameters corresponding to the inverted element excitation and the elevation plane sidelobe level in the pattern parameters before failure is within the allowable error of the sidelobe level.

Images