CN-121997479-A - Intelligent aided design method and device for heating system based on pre-training large model

CN121997479ACN 121997479 ACN121997479 ACN 121997479ACN-121997479-A

Abstract

The embodiment of the invention discloses an intelligent aided design method and device of a heating system based on a pre-training large model, and relates to the field of engineering design. The user puts forward knowledge problems about the heating system such as composition module design parameter data and the like at the client side, and the relevant heating system problem analysis is carried out at the server side. Firstly, a heating system module and system global knowledge data are acquired, a training sample database is constructed, a large language model is trained, answers of corresponding questions are analyzed by utilizing the trained large language model according to heating system related knowledge questions to be answered provided by a client, the answers aiming at the questions can be matched from the module database, the answers of the corresponding questions and the modules are finally returned to the client, engineers are assisted to design a heating system on the basis, the artificial workload is reduced, and meanwhile, a module question-answer result with higher accuracy is provided, and the working efficiency is improved.

Inventors

LIU XUEJUN
CAI XIN
WANG QI
WANG HEYUN
GUO YAN
WANG BO
LV HONGQIANG

Assignees

南京航空航天大学
青岛昊特智联能源有限公司
青岛凯能环保科技股份有限公司

Dates

Publication Date: 20260508
Application Date: 20251208

Claims (9)

1. An intelligent aided design method of a heating system based on a pre-training large model is characterized by comprising the following steps: s1, a server receives knowledge data and establishes a training sample database, wherein the knowledge data comprises module knowledge data of a heating system and global knowledge data of the heating system, and the training sample database is used for training a large language model; S2, receiving the heating system problems to be processed uploaded by the questioning terminal, and analyzing by using the large language model to obtain an analysis result; And S3, extracting a corresponding heating system module from the heating system module database according to the analysis result, and then feeding back the extracted heating system module and answer information to the questioning terminal.
2. The method of claim 1, wherein S1 comprises: According to the respective corresponding problem contents of the module knowledge data of the heating system and the global knowledge data of the heating system, the system is composed In the form of training data, the data is stored, In order to solve the problem, it is desirable, Answers to the related questions; The sample database further comprises working condition data of the heating system, wherein the working condition data comprise acquired temperature difference data and heat load data of the heating system.
3. The method of claim 1, wherein in training the large language model, comprising: and LoRA fine tuning process by utilizing module knowledge data of the heating system.
4. A method according to claim 2 or 3, characterized in that in training a large language model, it further comprises: and a Freeze fine tuning process utilizing global knowledge data of the heating system.
5. The method of claim 4, wherein the LoRA trim process utilizing module knowledge data of the heating system comprises: Freezing all layers of the large language model and training for newly added adapter parameters, wherein, , , A, B is an incremental weight used in approximate expression full parameter fine tuning Is initialized by Gaussian, matrix A is initialized by zero, and the quantity of the fine tuning parameters is adjusted from that of matrix B Is reduced to D represents d-dimensional parameters, r is rank, a is super-parameter, R represents A d-dimensional matrix of the matrix is provided, Representation d An r-dimensional matrix is provided which is a matrix, The input of the model is represented as such, The method comprises the steps of representing model output, gradient updating matrixes A and B in the process of training newly added adapter parameters, and weight merging in the process of reasoning and deployment.
6. The method of claim 5, wherein the Freeze fine tuning process utilizing global knowledge data of the heating system comprises: freezing the base layer of the large language model and updating parameters of the last 5 layers of the large language model, wherein the frozen parameters are as follows I represents the index of the parameter, k represents the boundary index for dividing the frozen layer and the trainable layer, i is less than or equal to k, and the gradient of the frozen parameter is reduced to zero in the process of updating the parameter; Parameters involved in fine tuning are J represents the index involved in the training update, j > k, and T is the number of iterations, t is, For learning rate, L is the loss function.
7. An intelligent aided design device of a heating system based on a pre-training large model is characterized by comprising: The data management module is used for receiving knowledge data and establishing a training sample database, wherein the knowledge data comprises module knowledge data of a heating system and global knowledge data of the heating system, and the training sample database is used for training a large language model; The analysis module is used for receiving the heating system problems to be processed uploaded by the questioning terminal and analyzing by utilizing the large language model to obtain analysis results; And the reply module is used for extracting the corresponding heating system module from the heating system module database according to the analysis result, and then feeding back the extracted heating system module and the answer information to the questioning terminal.
8. The apparatus according to claim 7, wherein the data management module is specifically configured to compose, based on the respective problem contents corresponding to the module knowledge data of the heating system and the global knowledge data of the heating system In the form of training data, the data is stored, In order to solve the problem, it is desirable, And the sample database also comprises working condition data of the heating system, wherein the working condition data comprises acquired temperature difference data and thermal load data of the heating system.
9. The apparatus of claim 7, wherein in training the large language model, comprising: a LoRA trim process using module knowledge data of the heating system, and a Freeze trim process using global knowledge data of the heating system.

Description

Intelligent aided design method and device for heating system based on pre-training large model Technical Field The invention relates to the technical field of engineering design of heating systems, in particular to an intelligent auxiliary design method and device of a heating system based on a pre-training large model. Background Conventional engineering flows face a number of challenges. On the one hand, product design often requires the combination of numerous parts into a complex whole, and manually designing the assembly relationships of these parts is not only time consuming but also prone to error. On the other hand, as the complexity of the product increases, the design effort in engineering drawing takes up a lot of time. In order to further improve the working efficiency of designers, artificial intelligence aided design tools based on large language models (Large Language Models, LLMs) are currently on-line. By combining the large language model with engineering design work, engineering design knowledge question and answer based on the large model is realized, and the optimization of design efficiency and assembly precision becomes an important direction of current research. This not only needs to solve the technical problem, but also needs to verify the feasibility and effectiveness thereof in practical application, thereby promoting further development of engineering design technology. While large models exhibit powerful natural language processing capabilities and change the vertical application paradigm for multiple domains, applications such as SMoE, chatGLM have not been widely used in industry. Therefore, the general corpus trained model of Bard et al cannot capture the proprietary expression of the industrial subdivision domain. In addition, the scarcity of the field data, the difficulty in acquiring high-quality labeling data in the professional field, the dependence of the model on general corpus training, and the deviation of the understanding of the professional terms, such as giving wrong or expired standard data of a heating system, are caused. These deviations occur during engineering, which can have very serious consequences, which can lead to inaccuracy and reliability problems in the application of large models in the field of industrial subdivision. Also, artificial intelligence aids, such as Deepseek, grox, which began to be applied on a large scale several years ago, were rapidly popular for some time and attempted to be applied to the engineering design field, but due to inaccuracy and credibility of answers given by the same, the subsequent auditing by experienced designers is often still required, and some design deviations are caused to bring about serious consequences, which results in that the current designers gradually give up such artificial intelligence tools on the design work of heating systems. Therefore, how to assist engineers in designing a heating system, while reducing the amount of human work, and providing highly accurate question-answering results, has become a subject to be studied. Disclosure of Invention The embodiment of the invention provides an intelligent auxiliary design method and device for a heating system based on a pre-training large model, which can assist engineers in designing the heating system, reduce the artificial workload and provide higher-accuracy question-answering results. In order to achieve the above purpose, the embodiment of the present invention adopts the following technical scheme: An intelligent aided design method of a heating system based on a pre-training large model comprises the following steps: s1, a server receives knowledge data and establishes a training sample database, wherein the knowledge data comprises module knowledge data of a heating system and global knowledge data of the heating system, and the training sample database is used for training a large language model; S2, receiving the heating system problems to be processed uploaded by the questioning terminal, and analyzing by using the large language model to obtain an analysis result; And S3, extracting a corresponding heating system module from the heating system module database according to the analysis result, and then feeding back the extracted heating system module and answer information to the questioning terminal. Specifically, S1 comprises the steps of forming according to the respective corresponding problem contents of the module knowledge data of the heating system and the global knowledge data of the heating systemThe sample database also comprises working condition data of the heating system, wherein the working condition data comprises acquired temperature difference data and heat load data of the heating system. For example, the knowledge data uploaded by the client comprises module knowledge data and system global knowledge data in the heating system and questions corresponding to the knowledge data, the training data is in a form of 'que