Data-driven scientific research space process modeling method
Through the data-driven scientific research space process modeling method, the hierarchical architecture and neural network algorithm are used to solve the synchronization and coupling problems of scientific research space business process modeling tools, realize the intelligent management of experimental resources and accurate results feedback, and improve scientific research Space operation efficiency and digital capabilities.
Patent Information
- Application Number
- CN202410918413.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-09
- Publication Date
- 2024-11-12
AI Technical Summary
The existing scientific research space business process modeling tools have problems such as untimely synchronization of view codes, high coupling of execution engines, and large model deviations, making it difficult to achieve efficient identification, tracking and management of experimental resources.
Using data-driven scientific research space process modeling method, the preprocessing, simulation, format conversion and actual experimental data are realized through hierarchical architecture (display layer, communication layer, business logic layer, algorithm layer, data layer) and neural network algorithm. Simulation, providing an open experimental environment, and intelligent management of experimental resources.
It realizes intelligent identification, tracking and management of experimental resources, provides accurate feedback on experimental results and equipment life cycle management, and improves the operation efficiency and digital capabilities of scientific research space.
Smart Images

Figure CN118940603A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of digital modeling of scientific research space business processes, and specifically to a data-driven scientific research space process modeling method. Background Art
[0002] The demand for business process expansion, reorganization and repositioning in modern scientific research space is increasing. With the continuous maturity of relevant technologies, it is necessary to model business processes when implementing a complete solution. Existing business process modeling tools are compared in terms of features, installation, ease of use, stability, etc., but there are still shortcomings such as untimely view code synchronization, excessive coupling with the execution engine, and large deviation of the built model. In recent years, in terms of algorithm models, with the development of machine learning, especially deep learning, a series of neural networks such as convolutional neural networks, recurrent networks, and generative adversarial networks have emerged, providing models and algorithms for the modeling and analysis of three-dimensional scenes. The quality of modeling has also been continuously improved thanks to the improvement of computer performance. In summary, with the rapid development of deep learning and the continuous expansion of computer memory and video memory, the research on process modeling and the development of three-dimensional modeling technology have obtained many convenient conditions, which also provide technical support for the process modeling of scientific research space.
[0003] The virtual modeling technology realized by computer technology provides a virtual laboratory environment for scientific research managers. The application of this virtual modeling technology provides the following real-time or quasi-real-time data from the basic data acquisition level, including space environment data, space environment device operation data and instrument operation data.
[0004] Therefore, the present invention proposes a data-driven scientific research space process modeling method for the real process of scientific research space and experimental site. Through this method, the intelligent identification, tracking, monitoring and management of experimental resources and the timely feedback of experimental process and results are realized, providing a completely open innovative experimental environment and practicing equipment life cycle management. To a certain extent, the operating efficiency of scientific research space is improved based on actual operational needs, and effective experimental data assets are formed based on experimental and operational management data to enhance digital capabilities. Summary of the invention
[0005] The purpose of the present invention is to provide a data-driven scientific research space process modeling method to solve the problems raised in the above background technology.
[0006] To achieve the above object, the present invention provides the following technical solution: a data-driven scientific research space process modeling method, comprising the steps of:
[0007] S1: Modeling architecture, building a data-driven scientific research space process modeling architecture. The scientific research space process modeling architecture is constructed in layers, including display layer, communication layer, business logic layer, algorithm layer, and data layer;
[0008] S2: Data import, interface configuration, experimental configuration data are saved through json files, running data is saved when needed during the experiment, and then resource data and experimental object 3D model data are stored in the data layer;
[0009] S3: Data algorithm processing, data is preprocessed, simulated, format converted, actually simulated, and digital-real fusion to complete the modeling of scientific research space process;
[0010] S4: Business logic is connected in series. According to the real scientific research scenarios, business modules are preset. The business modules include experiment management, interface management, resource management, scene management, and scene interaction. The business logic layer calls S1-S3 steps for users to complete the corresponding requests, and then feeds the results back to the display layer for display;
[0011] S5: Data transmission communication, decoupling each module and completing information transmission between modules through the message broadcast center and EventSystem module of the communication layer;
[0012] S6: Scientific research space process modeling and display, displaying the experimental process on the display layer, including modeling virtual scenes and graphical modeling interfaces for experimental data, forwarding user input requests to the business logic layer through the EventSystem module to complete related business logic, and interacting with step S4.
[0013] Preferably, the data layer and algorithm layer in the modeling architecture in S1 are called by the business logic layer, and the business logic layer and the display layer interact through the communication layer. At the same time, the display layer can call the business logic layer, but the business logic layer has no direct relationship with the display layer.
[0014] Preferably, the data algorithm processing of S3 specifically includes:
[0015] Data preprocessing: preprocess the provided data, remove abnormal data, reduce the effective interval, and convert the data structure into a structure that is easy to calculate. Classify the data of each scene, make a data set, and divide the data set into a training set and a validation set;
[0016] For simulation calculation, for the data of a given scenario, simple algorithms such as regression analysis are first used to process simple data on the boundary, and then neural network methods are used for overall processing and regression analysis;
[0017] Format conversion, converting digital simulation data into file data that can be accessed to the operation perspective layer, including log files or csv files;
[0018] In actual simulation, the results obtained after data processing in the logical perspective are connected to the operational perspective, and all the above steps are repeated to complete data processing and access in each scene to complete the modeling of the scientific research space process.
[0019] Preferably, in the process of digital simulation of a given scene, the boundary conditions need to be determined first. At this time, the regression analysis method needs to be used to fit the boundary data points to obtain the boundary conditions u=f(x) represented by the functional relationship. The boundary conditions will be the basis for the next step of digital simulation. Using the neural network training data, during the simulation process, we use the training set T to verify the performance, continuously optimize the output results of the model, and use the verification set V to verify whether the training results of the training set are correct. When the optimization reaches a certain level, the training results can be used to output the numerical results of the digital simulation.
[0020] Preferably, the message broadcast center in S5 is designed by the observer mode, and each module can monitor the event by subscribing to the event in the message broadcast center.
[0021] Compared with the prior art, the present invention has the following beneficial effects:
[0022] 1. The present invention proposes a data-driven scientific research space process modeling method for the real process of scientific research space and experimental site. This method can realize the intelligent identification, tracking, monitoring and management of experimental resources and timely feedback of experimental process and results, provide a completely open innovative experimental environment, and practice equipment life cycle management.
[0023] 2. The present invention constructs a data-driven scientific research space process modeling architecture, uses neural networks to extract high-dimensional features of three-dimensional target objects, and constructs a depth map generation network based on perspective fusion.
[0024] 3. The present invention uses a neural network in a data-driven manner to extract more abstract and more general features, and uses this high-dimensional feature as the basis for retrieval, so as to retrieve more similar models in the database and complete the modeling.
[0025] 4. The algorithm of the present invention can obtain the numerical simulation results of the actual operation data more accurately. In order to overcome the problem that the initial random point selection of the algorithm is particularly unreasonable and thus cannot converge to the actual operation data, the present application will reduce the learning rate as much as possible in the preprocessing stage, first slowly reduce the error to a range that can be quickly iterated, and then conduct optimization experiments, and finally complete the process of spatial modeling of scientific research sites through fusion.
[0026] 5. The innovation of the method of the present invention lies in the clear and distinct architecture. The system construction is based on a simulation model rather than a pure data model, and the hierarchy is clear and rich. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 This is a deep learning modeling flow chart in an embodiment of the present invention;
[0028] Figure 2 It is the overall architecture diagram of the present invention. DETAILED DESCRIPTION
[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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.
[0030] See also Figure 1-2 The present invention provides a technical solution: a data-driven scientific research space process modeling method, comprising the steps of:
[0031] S1: Modeling architecture, building a data-driven scientific research space process modeling architecture. The scientific research space process modeling architecture is constructed in layers, including display layer, communication layer, business logic layer, algorithm layer, and data layer;
[0032] S2: Data import, interface configuration, experimental configuration data are saved through json files, running data is saved when needed during the experiment, and then resource data and experimental object 3D model data are stored in the data layer;
[0033] S3: Data algorithm processing, data is preprocessed, simulated, format converted, actually simulated, and digital-real fusion to complete the modeling of scientific research space process;
[0034] S4: Business logic is connected in series. According to the real scientific research scenarios, business modules are preset. The business modules include experiment management, interface management, resource management, scene management, and scene interaction. The business logic layer calls S1-S3 steps for users to complete the corresponding requests, and then feeds the results back to the display layer for display;
[0035] S5: Data transmission communication, decoupling each module and completing information transmission between modules through the message broadcast center and EventSystem module of the communication layer;
[0036] S6: Scientific research space process modeling and display, displaying the experimental process on the display layer, including modeling virtual scenes and graphical modeling interfaces for experimental data, forwarding user input requests to the business logic layer through the EventSystem module to complete related business logic, and interacting with step S4.
[0037] In this embodiment, the data layer and algorithm layer in the modeling architecture in S1 are called by the business logic layer, and the business logic layer and the display layer both interact through the communication layer. At the same time, the display layer can call the business logic layer, but the business logic layer has no direct relationship with the display layer.
[0038] In this embodiment, the data algorithm processing of S3 specifically includes:
[0039] Data preprocessing: preprocess the provided data, remove abnormal data, reduce the effective interval, and convert the data structure into a structure that is easy to calculate. Classify the data of each scene, make a data set, and divide the data set into a training set and a validation set;
[0040] For simulation calculation, for the data of a given scenario, simple algorithms such as regression analysis are first used to process simple data on the boundary, and then neural network methods are used for overall processing and regression analysis;
[0041] Format conversion, converting digital simulation data into file data that can be accessed to the operation perspective layer, including log files or csv files;
[0042] In actual simulation, the results obtained after data processing in the logical perspective are connected to the operational perspective, and all the above steps are repeated to complete data processing and access in each scene to complete the modeling of the scientific research space process.
[0043] In this embodiment, during the digital simulation of a given scene, the boundary conditions must first be determined. At this time, a regression analysis method is needed to fit the boundary data points to obtain the boundary conditions u=f(x) represented by the functional relationship. The boundary conditions will be the basis for the next step of digital simulation. Using the neural network training data, during the simulation process, we use the training set T to verify the performance, continuously optimize the output results of the model, and use the verification set V to verify whether the training results of the training set are correct. When the optimization reaches a certain level, the training results can be used to output the numerical results of the digital simulation.
[0044] In the neural network method, the residual neural network method is used to solve the optimization problem, generate numerical solutions of digital simulation from the numerical values in the scene, and make it as close as possible to the various numerical values in the real scene to complete the final modeling process. In the residual neural network, the stochastic gradient descent method is generally used. The stochastic gradient descent method updates the weights and thresholds in the neural network by taking partial derivatives.
[0045] For the number of iterations t in the deep neural network, the threshold b t , the loss term L m , and a given number of random points N, the formula for the stochastic gradient descent method is as follows:
[0046]
[0047] In the neural network method, the above stochastic gradient descent method is first used to update the parameters. After the parameters are updated, iteration is performed based on the parameters. Taking a two-layer neural network as an example, the iteration formula is as follows:
[0048] T=σ(WX+B)
[0049] K=σ(WT+B)
[0050] Y=K+X
[0051] Where W is the weight, Composed vector, B is the threshold b t A vector consisting of X as input and Y as output.
[0052] The key to this method is: should the number of random points selected be relatively large or slightly small; should the learning rate η be large or small during the gradient descent process.
[0053] The algorithm can obtain the numerical simulation results of the actual operation data more accurately. In order to overcome the problem that the initial random point selection of the algorithm is particularly unreasonable and thus cannot converge to the actual operation data, this paper will reduce the learning rate as much as possible in the preprocessing stage. First, the error is slowly reduced to a range that can be quickly iterated, and then the optimization experiment is carried out. Finally, the process of spatial modeling of scientific research sites is completed through fusion.
[0054] Using the neural network training data, during the simulation process, we use the training set T to verify the performance, continuously optimize the output results of the model, and use the verification set V to verify whether the training results of the training set are correct. After 10,000 iterations in the training phase, the learning rate of the network is set to 0.0002, and two sets of data are sent as the input of the training set during each training process. This training process took 48 hours. The network structure is built on the Pytorch platform, the code is written in Python, and the network training process is carried out on the Nvidia Titan graphics card. When the optimization reaches a certain level, the results obtained from the training can be used to output the numerical results of the digital simulation.
[0055] In this embodiment, the message broadcast center in S5 is designed by the observer mode, and each module can monitor the event by subscribing to the event in the message broadcast center.
[0056] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A data-driven scientific research space process modeling method, characterized in that: Includes steps: S1: Modeling architecture, building a data-driven scientific research space process modeling architecture. The scientific research space process modeling architecture is constructed in layers, including display layer, communication layer, business logic layer, algorithm layer, and data layer; S2: Data import, interface configuration, experimental configuration data are saved through json files, running data is saved when needed during the experiment, and then resource data and experimental object 3D model data are stored in the data layer; S3: Data algorithm processing, data is preprocessed, simulated, format converted, actually simulated, and digital-real fusion to complete the modeling of scientific research space process; S4: Business logic is connected in series. According to the real scientific research scenarios, business modules are preset. The business modules include experiment management, interface management, resource management, scene management, and scene interaction. The business logic layer calls S1-S3 steps for users to complete the corresponding requests, and then feeds the results back to the display layer for display; S5: Data transmission communication, decoupling each module and completing information transmission between modules through the message broadcast center and EventSystem module of the communication layer; S6: Scientific research space process modeling and display, displaying the experimental process on the display layer, including modeling virtual scenes and graphical modeling interfaces for experimental data, forwarding user input requests to the business logic layer through the EventSystem module to complete related business logic, and interacting with step S4.
2. A data-driven scientific research space process modeling method according to claim 1, characterized in that: The data layer and algorithm layer in the modeling architecture of S1 are called by the business logic layer. The business logic layer and the display layer interact through the communication layer. At the same time, the display layer can call the business logic layer, but the business logic layer has no direct relationship with the display layer.
3. A data-driven scientific research space process modeling method according to claim 1, characterized in that: The data algorithm processing of S3 specifically includes: Data preprocessing: preprocess the provided data, remove abnormal data, reduce the effective interval, and convert the data structure into a structure that is easy to calculate. Classify the data of each scene, make a data set, and divide the data set into a training set and a validation set; For simulation calculation, for the data of a given scenario, simple algorithms such as regression analysis are first used to process simple data on the boundary, and then neural network methods are used for overall processing and regression analysis; Format conversion, converting digital simulation data into file data that can be accessed to the operation perspective layer, including log files or csv files; In actual simulation, the results obtained after data processing in the logical perspective are connected to the operational perspective, and all the above steps are repeated to complete data processing and access in each scene to complete the modeling of the scientific research space process.
4. A data-driven scientific research space process modeling method according to claim 3, characterized in that: In the process of digitally simulating a given scene, the boundary conditions need to be determined first. At this time, a regression analysis method needs to be used to fit the boundary data points to obtain the boundary conditions u=f(x) represented by the functional relationship.
5. A data-driven scientific research space process modeling method according to claim 1, characterized in that: The message broadcast center in S5 is designed by the observer mode, and each module can monitor the event by subscribing to the event in the message broadcast center.
Citation Information
Patent Citations
Visual industrial process management and control platform based on modeling technology
CN112528525A
Digital twinning-based sanitary pottery product assembly production management system and method
CN113344505A
Digital twin production line visualization system based on real-time data driving
CN116415386A
Chemical experiment monitoring optimization method and device based on digital twinning
CN116682500A
