CN-122021849-A - Data processing method and device applied to telescope observation scheduling

Abstract

The application provides a data processing method applied to telescope observation scheduling, and relates to the astronomical observation field and the artificial intelligence field. The method comprises the steps of obtaining first output data obtained when a large model takes user input demand information as input, wherein the first output data comprises task suggestion content generated by the large model, reasoning logic information of the large model and reference data, the user input demand information comprises information for indicating the large model to generate task suggestions of telescope observation scheduling tasks, executing data consistency check based on the task suggestion content and the reference data, matching reasoning logic information with reasoning logic rules in a preset expert knowledge base if the data consistency check result is that the data consistency check is passed, and classifying the first output data as trusted output if the reasoning logic rules matched with the reasoning logic information exist, otherwise classifying the first output data as to-be-confirmed output. The application also provides a data processing device applied to telescope observation scheduling.

Inventors

• Xiao Hengchu
• WANG CHUANJUN
• REN SIYING
• HE SHOUSHENG
• WANG ZHENZHEN
• FENG YUANJIE
• WANG CHUNPING

Assignees

• 中国科学院云南天文台

Dates

Publication Date

20260512

Application Date

20251230

Claims (15)

1. 1. A data processing method applied to telescope observation scheduling, comprising: acquiring first output data obtained when input demand information of a user is taken as input of a large model, wherein the first output data comprises task suggestion content generated by the large model, reasoning logic information and reference data of the large model; Performing a data consistency check based on the task suggestion content and the reference data; if the data consistency check result is that the check is passed, matching the reasoning logic information with the reasoning logic rules in a preset expert knowledge base; And if the inference logic rules matched with the inference logic information exist, classifying the first output data as trusted output, otherwise classifying the first output data as output to be confirmed.
2. 2. The method of claim 1, wherein the method further comprises: And when the first output data is classified as a trusted output, performing task scheduling based on task suggestion content in the first output data.
3. 3. The method of claim 1, wherein the performing a data consistency check based on the task suggestion content and the reference data comprises: performing parameter value rationality verification on the task suggestion content based on a preset verification rule to obtain a rationality verification result; performing database retrieval verification based on a preset database, performing authenticity verification on the reference data to obtain an authenticity verification result, and And when the rationality verification result and the authenticity verification result are both passed, determining that the data consistency verification result is verification passing.
4. 4. The method of claim 3, wherein the task suggestion content includes a pointing parameter, an imaging parameter, a time parameter, and an environmental parameter, the performing parameter value rationality verification on the task suggestion content based on a preset verification rule, and obtaining a rationality verification result includes: Verifying whether the pointing parameter is in a preset pointing constraint interval or not to obtain a first verification result; verifying whether the imaging parameters are within a preset imaging configuration allowable range or not to obtain a second verification result; Verifying whether the time parameter is within a preset time arrangement allowable range or not to obtain a third verification result; verifying whether the environmental parameter is within the preset environmental safety threshold value to obtain a fourth verification result, and And when the first verification result, the second verification result, the third verification result and the fourth verification result are all yes, determining that the task suggestion content passes the parameter value rationality verification.
5. 5. A method according to claim 3, wherein said verifying the authenticity of the reference data based on a preset database, obtaining an authenticity verification result comprises: inquiring to confirm whether the real data conforming to the reference data exists in the preset database or not, and And when the authenticity data is queried, determining that the authenticity verification of the reference data is passed.
6. 6. The method of claim 1, wherein the method further comprises: When the first output data is classified as to-be-confirmed output, requesting to perform user confirmation operation on the first output data; And obtaining a processing result of the user confirmation operation.
7. 7. The method according to claim 6, wherein after acquiring the processing result of the user confirmation operation, the method further comprises: And if the processing result comprises second output data which is processed to be credible by the user confirmation operation, performing task scheduling based on task suggestion content in the second output data, wherein, When the user confirms that the operation directly processes the first output data into credible, the second output data is the first output data; And when the user confirms that the operation modifies the first output data and processes the modified data as trusted, the second output data is the modified data.
8. 8. The method of claim 7, wherein the method further comprises: Carrying out structural extraction on the reasoning logic information in the second output data to generate a newly added reasoning logic rule; And storing the newly added reasoning logic rules to the expert knowledge base.
9. 9. The method of claim 8, wherein the method further comprises: periodically counting hit frequencies of each reasoning logic rule in the expert knowledge base in a history matching process; converting the reasoning logic rule with hit frequency exceeding a preset threshold value into structured prompt information; And introducing the prompt information into the user input demand information before the user input demand information is input into the large model, so that the large model obtains the first output data based on the user input demand information and the prompt information.
10. 10. The method of claim 8, wherein the method further comprises: periodically constructing a first training sample set based on reasoning logic rules in the expert knowledge base; And performing supervision fine tuning on the large model by using the first training sample set.
11. 11. The method of claim 7, wherein the method further comprises: constructing a second training sample set, and Based on the second training sample set, performing preference optimization training on the large model by adopting a reinforcement learning method; Wherein constructing the second training sample set comprises: and when the processing result comprises second output data which is processed to be credible by the user confirmation operation and is data obtained after the user modifies the first output data, taking the second output data as positive samples in the second training sample set and taking the first output data as negative samples in the second training sample set.
12. 12. A data processing apparatus for use in telescope observation scheduling, the apparatus comprising: The system comprises an output data acquisition module, a telescope observation scheduling task generation module and a telescope observation scheduling module, wherein the output data acquisition module is used for acquiring first output data obtained when a large model takes user input demand information as input, and the first output data comprises task suggestion content generated by the large model, reasoning logic information and reference data of the large model; A data consistency check module for performing data consistency check based on the task suggestion content and the reference data; the reasoning logic rule matching module is used for matching the reasoning logic information with the reasoning logic rules in a preset expert knowledge base if the data consistency check result is that the data consistency check result is passed; and the output data classification module is used for classifying the first output data into trusted output if the inference logic rule matched with the inference logic information exists, and classifying the first output data into output to be confirmed if the inference logic rule matched with the inference logic information exists.
13. 13. An electronic device, comprising: one or more processors; A memory for storing one or more computer programs, Characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1-11.
14. 14. A computer-readable storage medium, on which a computer program or instructions is stored, which, when executed by a processor, carries out the steps of the method according to any one of claims 1 to 11.
15. 15. A computer program product comprising a computer program or instructions which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 11.

Description

Data processing method and device applied to telescope observation scheduling Technical Field The application relates to the astronomical observation field and the artificial intelligence field, in particular to a data processing method, a device, equipment, a medium and a program product applied to telescope observation scheduling. Background The development of artificial intelligence technology promotes the scheduling control of telescopes and telescope arrays to gradually introduce large models for generating observation task suggestions and assisting in forming observation plans, the output quality is usually improved through the modes of field data training, model fine tuning, reinforcement learning and the like in the prior art, but because the large models are still possible to generate 'illusions' based on probability generation mechanisms, the situation that task suggestion content and reference data thereof are unreal, inconsistent or incompatible with scheduling constraint exists, telescope observation scheduling belongs to a high-reliability requirement scene, the situation can further bring risks of task arrangement conflict, equipment configuration mismatch or suggested execution when safety conditions are not met, meanwhile, the observation scheduling often involves the coexistence of multi-source heterogeneous data and multi-dimensional constraint, and in engineering landing, the aspects of interpretability, verifiability, consistency control of inference logic and existing experience rules and the like of large model output still remain to be perfected. Disclosure of Invention In view of the foregoing, the present application provides a data processing method, apparatus, device, medium, and program product for telescope observation scheduling. According to the first aspect of the application, a data processing method applied to telescope observation scheduling is provided, and the method comprises the steps of obtaining first output data obtained when a large model takes user input demand information as input, wherein the first output data comprises task suggestion content generated by the large model, reasoning logic information of the large model and reference data, the user input demand information comprises information for indicating the large model to generate task suggestions of telescope observation scheduling tasks, performing data consistency check based on the task suggestion content and the reference data, matching the reasoning logic information with reasoning logic rules in a preset expert knowledge base if the data consistency check result is that the data consistency check is passed, classifying the first output data as trusted output if the reasoning logic rules matched with the reasoning logic information exist, and classifying the first output data as to-be-confirmed output if the reasoning logic rules matched with the reasoning logic information exist. According to an embodiment of the present application, the data processing method applied to telescope observation scheduling further includes performing task scheduling based on task suggestion content in the first output data when the first output data is classified as a trusted output. According to the embodiment of the application, the data consistency verification is performed based on the task suggestion content and the reference data, wherein the data consistency verification comprises the steps of performing parameter value rationality verification on the task suggestion content based on a preset verification rule to obtain a rationality verification result, performing database retrieval verification on the reference data based on a preset database to obtain an authenticity verification result, and determining that the result of the data consistency verification is verification passing when both the rationality verification result and the authenticity verification result are passed. According to the embodiment of the application, the task suggestion content comprises a pointing parameter, an imaging parameter, a time parameter and an environment parameter, the parameter value of the task suggestion content is reasonably verified based on a preset verification rule, and the obtaining of the rationality verification result comprises the steps of verifying whether the pointing parameter is in a preset pointing constraint interval to obtain a first verification result, verifying whether the imaging parameter is in a preset imaging configuration permission range to obtain a second verification result, verifying whether the time parameter is in a preset time arrangement permission range to obtain a third verification result, verifying whether the environment parameter is in a preset environment safety threshold range to obtain a fourth verification result, and determining that the parameter value of the task suggestion content passes the rationality verification when the first verification result, the second verification