Method for analyzing data in aerospace operation and control field based on user portraits

Abstract

The invention discloses an aerospace operation and control field data analysis method based on a user portrait. The method comprises the steps of obtaining document data related to the aerospace operation and control field and storing the document data as structured data and semi-structured data; constructing a satellite user portrait for the structured data; constructing a user portrait of the task target for the semi-structured data; encoding each satellite operation and control task into a vector by adopting a position encoding algorithm, inputting the vectors into six continuous encoder layers of Transformer, and constructing and training a prediction model; extracting a task description tag from a task demand text provided by a user by using a named entity recognition technology, inputting the task description tag into the prediction model to obtain a satellite identifier list most possibly meeting the task demand, and comparing the satellite identifier list with a user portrait of a satellite and a user portrait of a task target to obtain a task target. And outputting the optimal satellite meeting the task requirement. The problems that operation control data in the aerospace operation control field are numerous and complicated, and task requirements are difficult to intelligently match can be solved.

Description

Technical Field

[0001] The present invention relates to the field of computer data analysis, and in particular, to a data analysis method, computer equipment, computer-readable storage medium and computer program product in the field of aerospace operation and control based on user portraits. Background Art

[0002] At present, satellite projects and commercial aerospace are developing rapidly. The measurement and control mode that focuses on manpower and material resources has been difficult to adapt to the current multi-task, multi-user, and multi-satellite ground operation and control service needs. Therefore, it is urgent to adopt an intelligent way to improve operation and control management and improve ground management efficiency. Satellite operation and control data is the key to building an intelligent data analysis system. As an existing technology for aerospace data analysis, CN116028566B discloses a spacecraft data analysis system, method, medium and equipment, and CN115291967B discloses a spacecraft data analysis method, device and electronic equipment. For satellite operation and control data, due to the wide variety of data, the data is complex, and there is a lack of a label system that clearly describes the requirements of mission satellites, it is difficult to independently complete resource allocation and task planning, which also limits the resource utilization and completion efficiency of operation and control tasks. Introducing an intelligent management and control system in the field of aerospace operation and control, it is urgent to have highly structured and effective satellite operation and control data, a data labeling system that is in line with user needs, and an accurate and efficient intelligent data analysis and retrieval method.

[0003] However, satellite data in the field of space operation and control is huge and difficult to label. There is relatively little research on the labeling of operation and control data, and the development is relatively slow. In this field, there are mainly problems such as insufficient user behavior information, high degree of specialization of domain terminology, and specialization of satellite demand labels, which leads to the complexity of knowledge in the field of space operation and control and the mixing of structured and semi-structured data (structured data is data with a clear data model and organizational structure, suitable for database system management; semi-structured data is data with a certain organizational structure and a certain degree of flexibility), making it difficult to accurately construct user features. The lack of an effective operation and control data and customer demand labeling system also makes it difficult for the intelligent analysis and retrieval of space operation and control data.

[0004] Secondly, the current satellite matching in the field of space operation and control mostly adopts manual methods, facing the increasingly complex all-weather space missions in the field of space ground operation and control, the surge in the number of spacecraft in orbit, and the increasingly complex multi-satellite application mode. At the same time, it has real-time and global coverage requirements in the fields of positioning navigation, information transmission, and reconnaissance observation, while the ground system may face problems such as equipment shortage, insufficient quantity, and high dependence on manual control. The existing ground system is difficult to adapt to the complex space-time layout and to meet the needs of various application functions, application scenarios, and application effects. When completing the task of geographical comprehensive situation analysis for the target location, it is necessary to collect a wide range of information with many types. These requirements will involve many sub-tasks, and require all-round data considerations to match the sub-tasks with large satellites of various functions and a large number of transiting small satellites. Since satellites undertake different observation tasks according to their payload type, flight cycle, orbital altitude and other attributes, the same ground station must have multi-satellite measurement and control and data collection capabilities to meet real-time user needs.

[0005] In summary, manual control solutions for space operation and control data are difficult to cope with the complexity of current space mission scenarios and the urgent need to match multi-angle and multi-faceted mission requirements of a large number of emerging satellite equipment. There is a lack of intelligent data analysis technology to maximize the utilization of existing equipment and significantly improve matching efficiency. Summary of the invention

[0006] The purpose of the present invention is to provide a data analysis method, computer equipment, computer-readable storage medium and computer program product in the field of aerospace operation and control based on user portraits, which can solve the problem that the operation and control data in the field of aerospace operation and control is complex and difficult to intelligently match with task requirements.

[0007] One aspect of the present invention provides a method for analyzing data in the field of aerospace control based on user portraits, comprising: Step S1, acquiring document data related to the field of space operation and control and storing them as structured data and semi-structured data, wherein the structured data adopts a multi-table joint index mechanism to form satellite operation and control mission records, and the semi-structured data is classified and stored from the perspective of user needs; Step S2: for structured data, count the frequency of occurrence of each satellite indicator in the satellite operation and control mission record, set a frequency threshold, filter out satellite indicators with a frequency exceeding the frequency threshold as preliminary labels, use an association rule algorithm to set a confidence threshold, filter out satellite indicators with a confidence higher than the confidence threshold that do not meet mission requirements as key labels, and construct a satellite user profile; for semi-structured data, use named entity recognition technology to extract entities and classify labels, and construct a user profile of the mission target; Step S3, using a position encoding algorithm to encode each satellite operation and control task in the satellite operation and control task record into a vector, and input the vector into six consecutive encoder layers of the Transformer, each encoder layer includes a self-attention sublayer and a feedforward neural network sublayer, the output of the self-attention sublayer is passed to the feedforward neural network, the output of the feedforward neural network is used as the final output of the encoder layer, and is passed to the next encoder layer, thereby constructing and training a prediction model; Step S4, using named entity recognition technology, according to the user profile of the task target, extracting task description tags from the task requirement text provided by the user, inputting the task description tags into the prediction model, obtaining a list of satellite identifiers that are most likely to meet the task requirements and a predicted probability value that may be competent for the task, and outputting the best satellite that meets the task requirements by comparing with the user profile of the satellite and the user profile of the task target.

[0008] Preferably, in step S2, entity extraction and label classification are performed as follows: For semi-structured data, BERT is used for pre-training and vectorization to perform entity extraction. The output of the last layer of BERT is used as the input of the bidirectional long short-term memory network layer for category prediction. The output of the bidirectional long short-term memory network layer is used as the input of the conditional random field layer to predict the label of each word in the document and classify it into predefined label categories.

[0009] Preferably, in step S3, the prediction model further includes a fully connected layer, which converts the output of the prediction model into a probability form through a Sigmoid activation function; The self-attention sublayer and feed-forward neural network sublayer of each encoder layer are subjected to residual connection and layer normalization operations.

[0010] Preferably, the constructed satellite user portrait includes the following types of tags: satellite identification, mission timeliness, roll angle, pitch angle, strip width, resolution, satellite type, and payload type.

[0011] Preferably, the user profile of the constructed task target includes the following types of tags: user code, equivalent resolution, thematic scenario, emergency participation, demand type, priority, observation probability, and time period preference.

[0012] Preferably, the satellite operation and control mission catalog includes the satellite identification, target longitude, target latitude, required start time, required end time, satellite resolution and mission results of each operation and control mission.

[0013] Another aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0014] Another aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0015] Another aspect of the present invention provides a computer program product, comprising a computer program, which implements the steps of the above method when executed by a processor.

[0016] The user portrait-based data analysis method, computer device, computer-readable storage medium and computer program product in the field of aerospace operation and control according to the above aspects of the present invention can solve the problem that the operation and control data in the field of aerospace operation and control is complex and difficult to match with task requirements intelligently. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings used in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work: Figure 1 This is a flow chart of a method for analyzing data in the field of aerospace control based on user portraits according to an embodiment of the present invention; Figure 2 4 is a structural diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. 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.

[0019] An embodiment of the present invention provides a data analysis method in the field of aerospace operation and control based on user portraits. The method of the embodiment of the present invention organically combines user portrait technology with data analysis technology based on deep learning, and can be effectively applied to data analysis and retrieval tasks in the field of aerospace operation and control. It can effectively represent different types of needs in detail through user portrait technology, and complete intelligent demand matching between satellite service centers and users, thereby improving the operation and control management capabilities and task completion efficiency of the service center.

[0020] Figure 1 FIG. 1 is a flow chart of a method for analyzing data in the field of aerospace control based on user portraits according to an embodiment of the present invention. Figure 1 As shown, the data analysis method in the field of aerospace operation and control based on user portraits according to an embodiment of the present invention includes steps S1 to S4.

[0021] Step S1 is a data processing step, in which document data related to the field of aerospace operation and control is acquired and stored as structured data and semi-structured data.

[0022] Specifically, various documents related to the field of space operation and control are obtained from websites and social media platforms such as Weibo and Zhihu, and valid documents are retained after document deduplication and data cleaning. After collecting and analyzing the main information of users in the above documents, such as various attributes, habits, and behavioral characteristics, the overall picture of user needs is summarized. The information extraction model is used to integrate the data, merge, clean and transform multi-source heterogeneous data, and store the collected massive satellite operation and control data as structured data and semi-structured data. The structured data adopts a multi-table joint index mechanism, with the requirement number (xqbh), meta-task number (yrwbh), satellite identifier (wxbs) as the primary key, to build a database and form a satellite operation and control mission record. The satellite operation and control mission record includes the satellite identifier, target longitude, target latitude, requirement start time, requirement end time, satellite resolution (resolution requirement) and mission results of each operation and control mission. The integrity and consistency of the data are guaranteed through the transaction isolation level. Semi-structured data is divided into six categories based on actual business scenarios and from the perspective of user needs, including peacetime needs, wartime needs, normal needs, emergency needs, basic needs and various new needs, and is stored in a classified manner.

[0023] Step S2 is the user portrait construction step. In this step, considering the complexity of data in the field of aerospace operation and control, as well as actual business needs, the construction of user portraits in the field of aerospace operation and control is divided into two parts, namely, the user portrait of the satellite and the user portrait of the mission target.

[0024] First, for the structured data obtained in step S1, satellite labels are constructed through data analysis and association rule mining technology, and labels describing satellite attributes and performance are screened by given frequency thresholds and confidence thresholds.

[0025] That is, for structured data, i.e., satellite operation and control mission records, the frequency of occurrence of each satellite indicator in the mission records is counted, a frequency threshold F is set, and satellite indicators with a threshold exceeding F are selected as preliminary labels. Secondly, focus on rejected cases in the operation and control records, i.e., cases where the inherent attributes or shooting tendencies of the satellite do not meet the mission requirements, and use association rule algorithms (such as FP-Growth frequent pattern tree) to set the confidence threshold G. For any satellite indicator A, if the confidence of the task rejection associated with A is higher than the confidence threshold G, the corresponding indicator is selected as the key label.

[0026] Secondly, for the semi-structured data obtained in step S1, a named entity recognition model is used to extract labels to describe the characteristics and requirements of the space operation and control mission objectives.

[0027] That is, for semi-structured data, combined with the actual needs of satellite operation and control tasks, an entity extraction model is used to extract labels. Named entity recognition technology (a technology that identifies named entities of specified types from text, such as names of people, places, and organizational names, and extracts them in a structured manner) is used to build an entity extraction model, which automatically completes the extraction of entities in documents and constructs labels.

[0028] Entity categories in the field of space operation and control include: space agencies, launch sites, space materials, astronautics, space missions, spacecraft types, spacecraft parameters, payload types, launch vehicle types, launch vehicle parameters, launch vehicle composition, ground equipment, space manufacturing, space technology, etc.

[0029] To address the difficulties in recognizing long entities and nested structures in aerospace texts, the BERT+Bi-LSTM+CRF (BBC) model is used for recognition.

[0030] First, the aerospace field text is pre-trained and vectorized through BERT (a natural language processing model) for representation layer to improve generalization ability. The vectorized text represents the contextual information within and between sentences, so as to accurately learn the characteristics of complex entities and improve the accuracy of complex entity recognition. The embodiment of the present invention uses a pre-trained module (using the Chinese model Chinese_L-12_H-768_A-12 based on BERT provided by Google) to capture inter-sentence semantics and identify nested entities. The output generated by the last layer of BERT is used as the input of the subsequent neural network model.

[0031] Secondly, the above output is used as the input of the bidirectional long short-term memory network (BiLSTM) layer to better solve the problem of long-distance dependency of aerospace text by evaluating the past and predicting the future functions. By training Bi-LSTM to learn each word in the long entity, category prediction is performed one by one, so as to more accurately express the actual meaning of a word in the long entity of the entire text. Finally, the output of the bidirectional long short-term memory network layer is used as the input of the conditional random field (CRF) layer to complete the model's prediction of the label of each word in the aerospace operation and control text and classify it into predefined categories. This combination of models effectively improves the accuracy of long entity recognition and complex entity recognition in the aerospace field. (1) Satellite User Profile The satellite user portrait constructed by the method of the embodiment of the present invention includes the following multiple tags: satellite identification, mission timeliness, roll angle, pitch angle, strip width, resolution, satellite type, and payload type.

[0032] Satellite identification: used to uniquely mark and distinguish a space satellite Product delivery / mission timeliness: This indicator is intended for qualitative (on time / delayed) analysis, counting the total number of on time / delayed missions of the satellite, and is used to evaluate the service stability of the product.

[0033] Roll angle: describes the roll amplitude of the satellite Pitch angle: describes the satellite pitch ·Strip width: An inherent property of the satellite, a range of values Resolution: The inherent property of the satellite camera lens. The digital code distinguishes different working modes. Satellite type: describes the satellite's function Payload type: Satellite payload mode The above indicators quantitatively represent the basic information of the satellite and are used to evaluate the performance and functional differences of the satellite.

[0034] The task user portrait constructed by the method of the embodiment of the present invention includes the following multiple tags: user code, equivalent resolution, thematic scene, emergency participation, demand grouping type, priority, observation probability, and time period preference.

[0035] User code: The unique identifier of the user who applies for satellite matching Equivalent resolution: The range of actual resolution required to observe the target. Thematic scenarios: Count the frequency of each scenario by demand source and use scenario codes to represent them.

[0036] Emergency involvement: The frequency of emergencies in the task.

[0037] Demand types: peacetime demand, wartime demand, normal demand, emergency demand, basic demand and various new demands Priority: Use a value range of 0-9999. The smaller the number, the higher the priority of the task.

[0038] Observation frequency: the frequency with which a target is photographed every day.

[0039] Time period preference: The time interval of intensive observation of a target is counted in units of ten minutes. Step S3 is a prediction model building step. In this step, the structured data in step S1 is used to extract satellite description labels based on the user portrait of the satellite in step S2, and used as training samples for the prediction model. An example of a training sample in an embodiment of the present invention is: satellite identification, target longitude and latitude, required start time, required end time, and resolution requirement for each operation and control mission.

[0040] The prediction model uses the ERNIE Tiny-Micro model to predict whether each satellite will accept or reject the user demand mission, and can output the three most likely optimal satellite identifiers based on the input user demand mission data. The ERNIE Tiny-Micro model is a small model obtained by distilling the ERNIE large model for satellite matching tasks in the field of space operation and control using large model distillation technology. Distillation is to transfer the knowledge of the large model to the small model so that the capabilities of the large model can be transferred to the small model. Since the size difference between the large model and the small model is more than a thousand times, model distillation is extremely difficult and even has the risk of failure. For this reason, the Wenxin ERNIE-Tiny distillation solution is used to introduce the teaching assistant model for distillation, and the teaching assistant is used as a bridge for knowledge transfer to solve the problem that the expression space of the small model and the large model is too far apart, thereby promoting the improvement of distillation efficiency.

[0041] The main component of the ERNIE Tiny-Micro model is the encoder layer of Transformer Blocks. Through training and learning of real data of mission requirements and satellite identification, the model automatically pushes the candidate satellite list. Structurally, the encoder contains 6 stacked blocks, each of which includes a self-attention mechanism and a feedforward neural network sublayer. In the prediction model of the embodiment of the present invention, each satellite mission description is calculated by the self-attention mechanism, and then further processed by the feedforward network to extract richer semantic information.

[0042] Use the embedding algorithm to encode each satellite mission in the database into a vector , the role of position encoding is to map discrete high-dimensional data into a low-dimensional dense continuous vector space in order to better represent and learn the characteristics of the data. is input into six consecutive encoder layers.

[0043]

[0044] The attention mechanism allows the input sequence to learn the relationship between each element, so that the model can better understand the meaning of the input sequence. In the attention sublayer of the input encoder layer, the corresponding The input vector is multiplied by the corresponding weight matrix (This matrix is ​​obtained by model training), calculate the Query, Key and Value corresponding to the vector, and get , , . Then the self-attention score (used to represent the relationship between an element and other elements in the sequence) is calculated. For any satellite mission vector , using the vector corresponding to With all the sequence Perform point multiplication to get the corresponding score. The multiplication of multiple vectors can be represented by multiplying matrices Q and K, that is, (in for The self-attention score amplitude can be further adjusted to make the neural network training more stable. The algorithm used in this module is to divide the obtained score by ,in Represents the dimension of vector k. Finally, the self-attention score obtained in the previous step is converted into a probability by a softmax operation. The score after softmax is multiplied by the corresponding Value, and then all Values ​​are added to form a new vector.

[0045]

[0046] A fully connected layer is added to integrate the information from the feedforward neural network and make the final classification decision. The feedforward neural network includes several linear transformations and nonlinear activation functions. The fully connected layer converts the output of the model into a probability form through the sigmoid activation function, where each dimension represents a category and the value ranges from 0 to 1, and x represents the output of the current sublayer of the model.

[0047]

[0048] Each sublayer of each encoder (Self-Attention layer and FFN layer) has a residual connection, and then performs a layer normalization operation. Sublayer(x) is a function implemented by the sublayer itself. Using LayerNorm, the output can be distributed near the origin, increasing sparsity and reducing the risk of overfitting, making it more suitable for the matching task of this project, that is, preventing the trained model from tending to output existing labels, resulting in a lower accuracy of matching results when facing more general satellite matching scenarios. This enhances the generalization and robustness of the model. The output of the residual and normalization layer is:

[0049] After the prediction model is trained, it will automatically match the most suitable satellite according to the input mission description, and output a list of satellite identifiers and the probability that they may be competent for the mission (i.e., the accuracy of the classification task). Use the top function to select the three mission satellites with the highest speculation probability (k=3). The model can output the optimal mission satellite list according to the input mission data. Among them, the sort function sorts the model output array v, and [1:k] refers to taking the first 1 to k elements of the array.

[0050]

[0051] Step S4 is a demand matching step. In this step, the task requirement text (i.e., observation order description text) provided by the user is matched with the requirements to obtain the corresponding task satellite recommendation.

[0052] Specifically, the named entity recognition technology in step S2 is used to extract the task description tag from the user's task requirement text according to the user portrait of the task target, and the task description tag is used as a piece of data that needs to be responded to. The user requirement data that needs to be responded to is input into the ERNIE Tiny-Micro prediction model to predict whether each satellite will choose to accept or reject the user requirement task, and the three most likely optimal satellite identifiers can be output according to the input task data. The user portrait can be manually compared according to the usage requirements, and a decision can be made based on reference standards such as the punctuality of the mission satellite.

[0053] The method of the embodiment of the present invention is described below by using a data example.

[0054] In step S1, the satellite operation and control mission record constructed by deduplication and cleaning of the acquired documents is as follows:

[0055] The satellite operation and control mission data in the database will be used to train satellite prediction models.

[0056] In step S2, the user profile of the satellite constructed by performing entity extraction on the acquired document is as follows:

[0057] The user profile of the task target is:

[0058] In step S3, a prediction model is constructed.

[0059] In step S4, a set of task description labels are extracted through named entity recognition:

[0060] The prediction model outputs the prediction results:

[0061] User portrait matching results:

[0062] In summary, the method of the embodiment of the present invention first obtains and integrates satellite operation and control documents to build a satellite operation and control mission database; secondly, the operation and control data documents are analyzed and entity labels are extracted, and user portraits of satellites and user portraits of mission targets are built according to user needs and historical preferences for satellite allocation; and the operation and control mission database is modeled and predicted using a deep learning model. The prediction model is used to screen and predict, and the three satellites that are most likely to meet user needs are output, and specific prediction probability values ​​are provided. Finally, the user portrait of the satellite, the user portrait of the mission target, and the intelligent prediction results of the model are comprehensively considered to output the best satellite that meets user needs and actual conditions. The method of the embodiment of the present invention makes full use of user portrait knowledge to extract satellite operation and control demand labels and deep learning models to complete intelligent analysis and retrieval of operation and control data, which effectively solves the problem that the operation and control data in the field of aerospace operation and control is complex and difficult to intelligently match task requirements.

[0063] An embodiment of the present invention further provides a computer device, which may be a server, and its internal structure diagram may be as shown in FIG. Figure 2As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store operating parameter data of each framework. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps of the method of the embodiment of the present invention are implemented.

[0064] Those skilled in the art will understand that Figure 2 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0065] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method of the embodiment of the present invention are implemented.

[0066] An embodiment of the present invention further provides a computer program product, including a computer program, which implements the steps of the method of the embodiment of the present invention when executed by a processor.

[0067] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A data analysis method in the field of aerospace control based on user portraits, characterized in that: include: Step S1, acquiring document data related to the field of space operation and control and storing them as structured data and semi-structured data, wherein the structured data adopts a multi-table joint index mechanism to form satellite operation and control mission records, and the semi-structured data is classified and stored from the perspective of user needs; Step S2: for structured data, count the frequency of occurrence of each satellite indicator in the satellite operation and control mission record, set a frequency threshold, filter out satellite indicators with a frequency exceeding the frequency threshold as preliminary labels, use an association rule algorithm to set a confidence threshold, filter out satellite indicators with a confidence higher than the confidence threshold that do not meet mission requirements as key labels, and build a satellite user profile; For semi-structured data, named entity recognition technology is used to extract entities and classify labels to build user portraits of task targets; Step S3, using a position encoding algorithm to encode each satellite operation and control task in the satellite operation and control task record into a vector, and input the vector into six consecutive encoder layers of the Transformer, each encoder layer includes a self-attention sublayer and a feedforward neural network sublayer, the output of the self-attention sublayer is passed to the feedforward neural network, the output of the feedforward neural network is used as the final output of the encoder layer, and is passed to the next encoder layer, thereby constructing and training a prediction model; Step S4, using named entity recognition technology, according to the user profile of the task target, extracting task description tags from the task requirement text provided by the user, inputting the task description tags into the prediction model, obtaining a list of satellite identifiers that are most likely to meet the task requirements and a predicted probability value that may be competent for the task, and outputting the best satellite that meets the task requirements by comparing with the user profile of the satellite and the user profile of the task target.

2. The method according to claim 1, characterized in that In step S2, entity extraction and label classification are performed as follows: For semi-structured data, BERT is used for pre-training and vectorization to perform entity extraction. The output of the last layer of BERT is used as the input of the bidirectional long short-term memory network layer for category prediction. The output of the bidirectional long short-term memory network layer is used as the input of the conditional random field layer to predict the label of each word in the document and classify it into predefined label categories.

3. The method according to claim 1 or 2, characterized in that In step S3, the prediction model further includes a fully connected layer, which converts the output of the prediction model into a probability form through a Sigmoid activation function; The self-attention sublayer and feed-forward neural network sublayer of each encoder layer are subjected to residual connection and layer normalization operations.

4. The method according to claim 1 or 2, characterized in that: The constructed satellite user portrait contains the following types of labels: satellite identification, mission timeliness, roll angle, pitch angle, swipe width, resolution, satellite type, and payload type.

5. The method according to claim 1 or 2, characterized in that: The user profile of the constructed task target includes the following types of labels: user code, equivalent resolution, thematic scenario, emergency participation, demand type, priority, observation probability, and time period preference.

6. The method according to claim 1 or 2, characterized in that: The satellite operation and control mission catalog includes the satellite identification, target longitude, target latitude, required start time, required end time, satellite resolution and mission results of each operation and control mission.

7. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images