CN-122021972-A - Method for generating substance scientific research experimental scheme based on AI model

CN122021972ACN 122021972 ACN122021972 ACN 122021972ACN-122021972-A

Abstract

The invention discloses a material scientific research experimental scheme generation method based on an AI model, and relates to the technical field of model cooperation. The method is based on a directed acyclic graph architecture, a plurality of different AI model executing nodes are called in parallel, the AI model executing nodes process scientific research questions simultaneously to generate a plurality of model answers, an aggregation node stores the model answers and corresponding session states, a break point is set in a call flow, the model answers are displayed to a user at the break point, one or more selected from the model answers by the user are received as selected answers, a summary node uses a preset fusion prompt word to guide the fusion AI model executing nodes to generate comprehensive answers based on the selected answers, and the problems that answer quality of an existing single large model or serial multi-model scheme is uneven, response efficiency is low, the user cannot participate in a decision and a multi-source result lacks an efficient fusion mechanism are solved.

Inventors

JIANG JUN
CHEN LINJIANG
LIU YUEBO
LIU SHIHONG

Assignees

中国科学技术大学

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (10)

1. The generation method of the material scientific research experimental scheme based on the AI model is characterized by comprising the following steps of: The starting node receives the user input scientific research problems; based on a directed acyclic graph architecture, a plurality of different AI model executing nodes are called in parallel, so that the AI model executing nodes process the scientific research questions simultaneously to generate a plurality of model answers; The sink node stores the model answers and the corresponding session states; Setting a break point in the call flow, displaying the model answer to a user at the break point, and receiving one or more selected by the user from the model answers as selected answers; and the summary node uses a preset fusion prompt word to guide the fusion AI model execution node to generate a comprehensive answer based on the selected answer as a final answer to the scientific research question.
2. The method for generating a substance scientific research experimental scheme based on an AI model according to claim 1, wherein the procedure of calling a plurality of different AI models in parallel based on a directed acyclic graph architecture is as follows: The starting node carries out pretreatment and information initialization on the scientific research problem; Triggering at least four AI model executing nodes to run in parallel by the starting node at the same time, and calling a corresponding large language model or knowledge base question-answering agent by each AI model executing node to generate an answer; and waiting for completion of all model execution nodes by the sink node, and converging and storing all generated answers.
3. The AI-model-based material scientific research experiment scheme generation method of claim 2, wherein the four AI-model execution nodes include DeepSeek model execution nodes, spark model execution nodes, scienceOne model execution nodes and knowledge base proxy execution nodes.
4. The method for generating a material scientific research experiment scheme based on an AI model as claimed in claim 3, wherein when the node is executed through the knowledge base agent to generate an answer, the scientific research questions after preprocessing and information initialization are queried in the knowledge base, and the content with the query score larger than the set score is used as the answer.
5. The method for generating a substance scientific research experimental scheme based on an AI model according to claim 3, wherein the DeepSeek model execution node, the Spark model execution node, and the ScienceOne model execution node all generate answers based on corresponding preset model prompt words.
6. The method for generating a substance scientific research experiment scheme based on an AI model of claim 3, wherein the knowledge base agent execution node generates an answer based on a preset knowledge base prompt word.
7. The method for generating a substance scientific research experiment scheme based on an AI model of claim 6, wherein the preset knowledge base prompt word enables the knowledge base agent to adopt a reference format of a numerical number in an answer, and provides a list of reference documents at the end of the answer.
8. The method for generating a substance scientific research experimental scheme based on an AI model as claimed in claim 1, wherein the preset fusion prompt word is used for guiding the fusion AI model execution node to extract information and views from the selected answers, identifying and merging repeated contents, and organizing information points into structured comprehensive answers according to logical relations.
9. The method for generating a material scientific research experiment scheme based on an AI model according to claim 1, wherein a history dialogue record related to a scientific research question is loaded after a comprehensive answer is generated, and a corresponding dialogue memory is constructed for each AI model execution node to be called.
10. The method for generating a substance scientific research experimental scheme based on an AI model of claim 9, wherein the process of constructing the corresponding dialogue memory is: Aiming at the history dialogue record, if the history comprehensive answers generated by fusion exist in the history dialogue record, distributing the history comprehensive answers to all AI model executing nodes as assistant reply contents; if the historical comprehensive answers do not exist, extracting original answers generated by the executing node in the historical dialogue record for each AI model as assistant reply contents; And combining the user problems in each history dialogue record with the assistant reply content distributed for the corresponding AI model execution node to form a corresponding dialogue memory.

Description

Method for generating substance scientific research experimental scheme based on AI model Technical Field The invention relates to the technical field of model cooperation, in particular to a material scientific research experimental scheme generation method based on an AI model. Background In the generation of a substance scientific research experimental scheme, the cooperation of multiple models to obtain comprehensive and accurate answers is a key link for improving the scientific research efficiency. However, in practical applications, the prior art solutions expose the following drawbacks: Firstly, the prior art relies on a few models which are called singly or serially, the knowledge range, the professional field and the reasoning logic of the prior art have inherent limitations, and the coupling requirements of complex scientific research problems in multiple dimensions such as material synthesis, characterization analysis, mechanism exploration, process optimization and the like cannot be covered, so that answers are one-sided and blind spots exist. Secondly, the existing scheme adopts a mode of model serial call, total consumption time is linearly increased, and a user is in a passive waiting state in the whole call process and cannot intervene in guiding, so that response efficiency is low and user experience is split. Finally, when answers to multiple models are obtained, the prior art lacks an effective mechanism capable of automatically fusing multi-source information, retaining core views, solving content conflicts and generating logical coherent comprehensive answers, and still relies on manual tedious comparison and integration. Therefore, a collaborative generation scheme which has the advantages of efficient parallel calling, supporting active feedback of users and realizing intelligent fusion by utilizing the self capacity of a large model is needed, so that the technical bottleneck is solved, and the quality, efficiency and controllability of the generation of a material scientific research experimental scheme are improved. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a material scientific research experimental scheme generation method based on an AI model, which solves the problems that the answer quality of the existing single large model or serial multi-model scheme in the scientific research question answer is uneven, the response efficiency is low, a user cannot participate in decision making and the multi-source result lacks an efficient fusion mechanism. The technical scheme of the invention for realizing the purpose is that the method for generating the experimental scheme of the material scientific research based on the AI model comprises the following steps: The starting node receives the user input scientific research problems; based on a directed acyclic graph architecture, a plurality of different AI model executing nodes are called in parallel, so that the AI model executing nodes process the scientific research questions simultaneously to generate a plurality of model answers; The sink node stores the model answers and the corresponding session states; Setting a break point in the call flow, displaying the model answer to a user at the break point, and receiving one or more selected by the user from the model answers as selected answers; and the summary node uses a preset fusion prompt word to guide the fusion AI model execution node to generate a comprehensive answer based on the selected answer as a final answer to the scientific research question. Further, the process of calling a plurality of different AI models in parallel based on the directed acyclic graph architecture is as follows: The starting node carries out pretreatment and information initialization on the scientific research problem; Triggering at least four AI model executing nodes to run in parallel by the starting node at the same time, and calling a corresponding large language model or knowledge base question-answering agent by each AI model executing node to generate an answer; and waiting for completion of all model execution nodes by the sink node, and converging and storing all generated answers. Further, the four AI model execution nodes include DeepSeek model execution nodes, spark model execution nodes, scienceOne model execution nodes, and knowledge base agent execution nodes. Further, when the node is executed through the knowledge base agent to generate an answer, the scientific research questions after preprocessing and information initialization are inquired in the knowledge base, and the content with the inquired score being larger than the set score is used as the answer. Further, the DeepSeek model execution node, the Spark model execution node and the ScienceOne model execution node all generate answers based on corresponding preset model prompt words. Further, the knowledge base agent executing node generates an answer based on a preset knowledg