AI modeling and data comparison method for chemical material characteristic analysis

Abstract

The invention relates to the technical field of chemical material characteristic analysis, and discloses an AI modeling and data comparison method for chemical material characteristic analysis. Characteristic parameters such as composition, physical form and the like of chemical materials are collected in real time through a multi-source sensing device and input into a preset feature extraction model to extract key feature vectors, and then multi-dimensional fitting is carried out through an AI modeling module according to material category relevance and experimental condition weights to generate a prediction model. And matching and verifying the prediction model and the real-time characteristic parameters by using a data comparison algorithm, and storing to form a dynamic analysis data set. The method further relates to feature extraction model construction, data processing, AI modeling step optimization, data classification and integration, authority control and the like. According to the method, efficient and accurate analysis of chemical material characteristics is realized, data processing and model prediction capabilities are improved, data safety is guaranteed, and the method has wide application prospects in the fields of chemical material research and production.

Description

Technical Field

[0001] The present invention relates to the technical field of chemical material property analysis, specifically an AI modeling and data comparison method for chemical material property analysis. Background Technique

[0002] In the field of chemical material research and application, accurately analyzing material properties is crucial for promoting the development of materials science, improving product quality, and optimizing production processes. Traditional chemical material property analysis methods have many limitations and are difficult to meet the needs of the rapid development of modern technology.

[0003] In the early stage, chemical material property analysis mainly relied on manual operations and simple experimental instruments. Researchers obtained material property data by manually performing chemical titrations, observing changes in the appearance of materials, etc. This method was extremely inefficient and greatly affected by human factors. Differences in the experimental techniques of different operators might lead to data deviations, and it was difficult to guarantee the accuracy and reliability of the data. With the progress of technology, some automated experimental equipment has been applied, which has improved the efficiency of data collection to a certain extent. However, these devices can often only measure single or a few properties and cannot comprehensively obtain information on the composition, physical form, thermodynamic properties, reaction activity indicators, etc. of chemical materials.

[0004] In the data processing and analysis section, traditional methods usually process the collected data based on simple statistical principles. Facing increasingly complex chemical material systems and massive experimental data, this method is difficult to uncover the deep - seated correlations and laws behind the data. For example, when studying new composite materials, their complex component combinations and interactions make it impossible for traditional data processing methods to accurately analyze the contributions of each component to the overall properties of the material and are also difficult to predict the performance changes of the material under different conditions.

[0005] In terms of modeling and prediction, most of the previous models were constructed based on empirical formulas and lacked a deep understanding of the essence of material properties. These models can often only be applied within specific experimental conditions and material ranges. Once the experimental conditions change or when facing new material systems, the prediction accuracy of the models will drop significantly. For example, when predicting the high - temperature mechanical properties of materials, traditional models are difficult to consider the coupling effects of multiple factors such as temperature, pressure, and changes in the microstructure of the material, resulting in a large deviation between the prediction results and the actual situation.

[0006] Existing AI-based analysis methods have deficiencies in aspects such as the comprehensiveness of data collection, the accuracy of feature extraction, and the rationality of model construction. For example, when dealing with multi-source heterogeneous data, some AI models cannot effectively integrate different types of sensor data, resulting in information loss; during the feature extraction process, they cannot dynamically adjust the extraction strategy according to the differences in material categories, affecting the recognition effect of key features; the constructed AI models are not comprehensive enough when considering the relevance of material categories and the weights of experimental conditions, limiting the prediction accuracy and generalization ability of the models. Summary of the Invention

[0007] The purpose of the present invention is to provide an AI modeling and data comparison method for analyzing the characteristics of chemical materials to solve the problems presented in the above background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solution: An AI modeling and data comparison method for analyzing the characteristics of chemical materials, the method includes: Real-time collection of characteristic parameters of chemical materials through multi-source sensing devices, wherein the characteristic parameters include composition, physical form, thermodynamic properties, and reaction activity indicators; Input the characteristic parameters into a preset feature extraction model to identify and extract key feature vectors, wherein the feature extraction model dynamically adjusts the extraction threshold based on the characteristics of different material categories in historical material data; Input the extracted key feature vectors into a preset AI modeling module to generate a material characteristic prediction model, wherein the AI modeling module performs multi-dimensional fitting according to the relevance of material categories and the weights of experimental conditions; Use a preset data comparison algorithm to match and verify the prediction model and real-time characteristic parameters, form a dynamic analysis data set and store it.

[0009] Preferably, the steps of constructing the feature extraction model include: Obtain a historical material data set, wherein each piece of data in the historical material data set is labeled with a material category and a characteristic level; divide training subsets based on the material category and characteristic level, and each training subset corresponds to a material scenario; Use the training subsets to parallel-train an initial extraction model until the feature recognition accuracy of the initial extraction model for each material scenario is greater than or equal to a preset first threshold, and then stop training to obtain an intermediate extraction model; Input the historical material data set into the intermediate extraction model, and verify whether the feature recognition results output by the intermediate extraction model meet the preset error range; if they meet, determine the intermediate extraction model as the feature extraction model.

[0010] Preferably, the real-time acquisition of the characteristic parameters of chemical materials by the multi-source sensing device includes: Establish a communication connection with the target sensing terminal, where the target sensing terminal is deployed at a preset monitoring point of the chemical material experimental device; Continuously read the real-time data of the target sensing terminal according to a preset acquisition frequency, and mark the acquisition position information based on the spatial distribution characteristics of the real-time data; According to the experimental process topology of the chemical material, physically align the real-time data at different monitoring points in the same experimental stage to form an associated characteristic parameter set.

[0011] Preferably, inputting the characteristic parameters into the preset feature extraction model includes: Extract the high-dimensional data segments in the characteristic parameters, where the high-dimensional data segments are data segments with parameter dimensions exceeding a preset dimensionality reduction threshold within a continuous acquisition period; Generate a feature screening index based on the dimensionality distribution and correlation of the high-dimensional data segments; Dynamically select the corresponding dimensionality reduction algorithm according to the feature screening index, where the sparse high-dimensional data uses the principal component dimensionality reduction algorithm, and the dense low-dimensional data uses the factorization algorithm.

[0012] Preferably, the method further includes: After extracting the key feature vectors, perform data integrity verification on the characteristic parameters; If it is found through verification that the data missing rate exceeds a preset second threshold, trigger the AI modeling module to interpolate and complete the missing data, where the high-priority missing data is the experimental parameter segment that affects the core characteristics.

[0013] Preferably, the AI modeling module includes the following modeling steps: Construct a dynamic weight model according to the experimental association network of the chemical material, where each experimental node corresponds to a conditional weight coefficient; Calculate the modeling collaborative weight based on the characteristic differences between adjacent experimental nodes; Combined with the experimental historical trend of the characteristic parameters, perform multi-dimensional predictive modeling on the unmeasured experimental nodes.

[0014] Preferably, the method further includes: After the modeling is completed, perform logical verification on the prediction model, where the verification method includes comparing the predicted characteristics with the measurement deviation of the actual experimental nodes; If the deviation exceeds a preset third threshold, readjust the modeling collaborative weight and iterate the modeling until the deviation is less than the third threshold.

[0015] Preferably, classifying and integrating the prediction model and real-time characteristic parameters by using a preset data comparison algorithm includes: Dividing first-level classification labels according to the type of experiment, where the first-level classification labels include basic physical property categories, reaction kinetics categories, and structural stability categories; Under each first-level classification label, further dividing second-level classification sub-labels based on the importance of the experiment; Storing the classified experimental data in different partitions of a distributed database according to the label hierarchy.

[0016] Preferably, the method further includes: Configuring the access level of the classification label according to a preset experiment permission; When receiving a data query request, verifying whether the permission identifier provided by the requester matches the access level of the target classification label; If they match, opening the data query channel corresponding to the classification label.

[0017] Preferably, the present invention further includes an AI modeling and data comparison electronic device for analyzing the characteristics of chemical materials, and the device includes: A multi-source sensing module for real-time collecting the characteristic parameters of chemical materials through multi-source sensing devices, where the characteristic parameters include composition, physical form, thermodynamic properties, and reaction activity indicators; A feature extraction module for inputting the characteristic parameters into a preset feature extraction model to identify and extract key feature vectors, where the feature extraction model dynamically adjusts the extraction threshold based on the characteristics of different material categories in historical material data; An AI modeling module for inputting the extracted key feature vectors into a preset AI modeling module to generate a material characteristic prediction model, where the AI modeling module performs multi-dimensional fitting according to the material category relevance and experimental condition weights; A data comparison module for using a preset data comparison algorithm to perform matching verification on the prediction model and real-time characteristic parameters, forming a dynamic analysis data set and storing it.

[0018] Compared with the prior art, the beneficial effects of the present invention are: In the data acquisition stage, multi-source sensing devices are used to collect real-time characteristic parameters such as the composition, physical form, thermodynamic properties, and reaction activity indicators of chemical materials. This method realizes the comprehensiveness and real-time nature of data acquisition. Compared with traditional methods that can only obtain single or a few types of characteristic data, it can provide richer and more accurate original data for subsequent analysis. For example, when studying new catalyst materials, multi-source sensing devices can simultaneously monitor the composition changes, surface morphology changes, temperature changes, and real-time reaction activity data of the catalyst during the reaction process, helping researchers comprehensively understand the characteristics of the catalyst at different stages, and thus providing a strong basis for optimizing the performance of the catalyst.

[0019] In the feature extraction stage, the preset feature extraction model dynamically adjusts the extraction threshold based on the characteristics of different material categories in historical material data. This adaptive feature extraction method can accurately identify and extract key feature vectors. Compared with traditional feature extraction methods with fixed thresholds, it can better adapt to the characteristics of different material categories, avoiding the omission or misjudgment of key features caused by unreasonable threshold settings. Taking metal materials and organic materials as examples, their characteristics are very different. This feature extraction model can dynamically adjust the extraction threshold according to the characteristics of each material category, accurately extract the key feature vectors reflecting their essential characteristics, and provide a high-quality data foundation for subsequent modeling.

[0020] The AI modeling module generates a material property prediction model through multi-dimensional fitting based on material category relevance and experimental condition weights, which greatly improves the accuracy and reliability of the model. In practical applications, there may be some potential correlations between different material categories, and at the same time, the influence of experimental conditions on material properties varies. By considering these factors, the model can more truly reflect the change law of material properties. For example, when studying the reaction characteristics of various materials under different temperature and pressure conditions, this AI modeling module can comprehensively analyze the correlations between material categories and the weights of different experimental conditions, accurately predict the reaction characteristics of materials under various complex conditions, and provide reliable theoretical guidance for experimental design and industrial production.

[0021] Use a preset data comparison algorithm to match and verify the prediction model and real-time characteristic parameters, and form a dynamic analysis data set for storage. This can not only timely detect the deviation between the model prediction result and the actual situation, and then optimize and improve the model, but also accumulate a large amount of valuable data for subsequent research and applications. For example, in the process of material quality inspection, by comparing the prediction model with the actually detected characteristic parameters in real time, once the deviation exceeds the preset range, the production process parameters can be adjusted in time to ensure the stability of product quality. At the same time, the long-term accumulated dynamic analysis data set can be used to further train and optimize the model, improve the generalization ability of the model, and enable it to better adapt to different experimental scenarios and material systems.

[0022] In addition, the present invention also has advantages in data management. By classifying and integrating experimental data and configuring the access levels of classification labels according to preset experimental permissions, it not only facilitates the storage, query, and management of data, but also ensures the security of data. In large-scale scientific research projects or industrial production, a large amount of experimental data needs to be managed in an orderly manner. The classified storage method enables researchers to quickly find the required data and improves work efficiency. At the same time, strict permission management ensures the security of sensitive data, prevents data leakage and abuse, and provides a strong guarantee for the smooth progress of scientific research and production activities. Brief Description of the Drawings

[0023] Figure 1 It is the working principle diagram of the AI modeling and data comparison method for chemical material characteristic analysis according to the present invention; Figure 2 It is the flowchart of the characteristic parameter input feature extraction model processing; Figure 3 It is the flowchart of the prediction model logic verification and adjustment; Figure 4 It is the flowchart of data access permission control. Detailed Embodiment

[0024] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the protection scope of the present invention.

[0025] Please refer to Figures 1-4 , the present invention provides an AI modeling and data comparison method for chemical material characteristic analysis, and the following will elaborate on its specific implementation.

[0026] Use multi-source sensing devices to collect the characteristic parameters of chemical materials in real time. These multi-source sensing devices can be various devices such as compositional analyzers, microscopes, thermal analyzers, reaction activity monitors, etc. The characteristic parameters include compositional composition, that is, the content of various chemical components in the chemical material; physical form, such as whether the material is solid, liquid or gaseous, and its shape, particle size, etc.; thermodynamic properties, such as specific heat capacity, thermal conductivity, melting point, etc.; reaction activity indicators, such as reaction rate constant, activation energy, etc. Through these multi-source sensing devices, real-time information of chemical materials can be obtained from multiple dimensions, providing a rich data basis for subsequent analysis.

[0027] Input the collected characteristic parameters into a preset feature extraction model. This feature extraction model is constructed based on historical material data, and it will dynamically adjust the extraction threshold according to the characteristics of different material categories in the historical material data. Different material categories have different characteristic distributions. For example, the compositional composition, physical form and other characteristics of metal materials and polymer materials are very different. The model will automatically adjust the extraction sensitivity to different characteristics according to these differences, so as to more accurately identify and extract the key feature vectors. These key feature vectors can represent the core characteristics of chemical materials and provide key information for subsequent modeling.

[0028] Input the extracted key feature vectors into a preset AI modeling module. The AI modeling module will perform multi-dimensional fitting according to the material category relevance and experimental condition weights. Different categories of chemical materials may have certain correlations in characteristics. For example, in some metal alloy materials, the change in compositional composition will simultaneously affect the physical form and thermodynamic properties. The experimental condition weight is set according to the importance of different experimental conditions on the material characteristics. For example, under high-temperature experimental conditions, the influence weight of temperature on the reaction activity of the material may be relatively large. By comprehensively considering these factors for multi-dimensional fitting, a model that can accurately predict the characteristics of chemical materials is generated.

[0029] Use a preset data comparison algorithm to match and verify the generated prediction model and the real-time collected characteristic parameters. The data comparison algorithm will compare the characteristic prediction values output by the prediction model with the actually collected characteristic parameters to check whether the difference between the two is within a reasonable range. After matching and verification, the relevant data is sorted to form a dynamic analysis data set and stored. These stored data can be used for subsequent analysis, research and further optimization of the model.

[0030] The technical solution of the present invention will be further described in detail below with specific embodiments.

[0031] Example 1: When constructing a feature extraction model, first obtain a historical material dataset. This dataset is the data accumulated from past experiments and analyses of various chemical materials, and each piece of data is labeled with the material category and property level. The material categories cover common metal materials, inorganic non-metallic materials, organic polymer materials, etc.; the property levels are divided into different levels according to certain criteria based on the various property indicators of the materials. For example, the strength property of the materials is divided into four levels: excellent, good, medium, and poor.

[0032] Based on the material category and property level, divide the training subsets, and each training subset corresponds to a material scenario. For example, for metal materials, the training subsets can be divided according to different metal types, such as iron-based alloys, aluminum-based alloys, etc.; for inorganic non-metallic materials, they can be divided according to different types such as ceramics and glass. After dividing the training subsets, use these training subsets to train the initial extraction model in parallel. The initial extraction model can adopt deep learning model architectures such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs). During the training process, continuously adjust the parameters of the model to continuously improve the feature recognition accuracy of the model for each material scenario. When the feature recognition accuracy of the model for each material scenario is greater than or equal to a preset first threshold (assuming the first threshold is 90%), stop the training to obtain an intermediate extraction model.

[0033] Input the historical material dataset into the intermediate extraction model to verify whether the feature recognition results output by the intermediate extraction model meet the preset error range. The preset error range can be set according to actual requirements and data precision requirements. For example, set the error range within ±5%. If it meets the preset error range, determine this intermediate extraction model as the final feature extraction model; if not, further adjust the model parameters or reselect the model architecture and conduct training and verification again.

[0034] Example 2: When real-time collecting the characteristic parameters of chemical materials through multi-source sensing devices, first establish a communication connection with the target sensing terminal. The target sensing terminal is deployed at the preset monitoring points of the chemical material experiment device, and these monitoring points are carefully selected according to the experimental requirements and characteristic distribution of the chemical materials. For example, when studying the heat conduction characteristics of materials, temperature sensors will be arranged at different parts of the materials as the target sensing terminal to monitor the temperature changes at different positions.

[0035] After establishing a communication connection, continuously read the real-time data of the target sensing terminal according to a preset acquisition frequency. The preset acquisition frequency is determined according to the requirements of the experiment and the speed of change of material characteristics. For example, for chemical materials with a relatively fast reaction rate, the acquisition frequency may be set to once per second; for materials with a relatively slow change, the acquisition frequency can be set to once per minute. While reading the real-time data, mark the acquisition position information based on the spatial distribution characteristics of the real-time data. For example, mark the acquisition position through the sensor number or the coordinate position on the experimental device, so as to facilitate subsequent data analysis and processing.

[0036] According to the experimental process topology of the chemical material, physically align the real-time data of different monitoring points in the same experimental stage. The experimental process topology describes the relationship between each stage and different monitoring points during the experiment. For example, in the synthesis experiment of materials, different stages involve the monitoring of parameters such as temperature, pressure, and composition. In a certain synthesis stage, organize and align the real-time data from different monitoring points such as temperature sensors and pressure sensors according to the logical relationship of the experimental process to form a set of associated characteristic parameters. This can ensure the integrity and consistency of the data and provide an accurate data basis for subsequent analysis.

[0037] Example 3: When inputting the collected characteristic parameters into a preset feature extraction model, first extract the high-dimensional data segments in the characteristic parameters. High-dimensional data segments refer to data segments with a parameter dimension exceeding a preset dimensionality reduction threshold within a continuous acquisition period. The preset dimensionality reduction threshold is determined according to the overall dimension of the data and the actual analysis requirements. Assuming the preset dimensionality reduction threshold is 10 dimensions, when the data dimension within a continuous acquisition period exceeds 10 dimensions, this data segment is a high-dimensional data segment.

[0038] Generate feature screening indicators based on the dimension distribution and correlation of the high-dimensional data segments. The dimension distribution reflects the change situation and importance of data in different dimensions, and the correlation reflects the mutual relationship between data in each dimension. For example, by calculating the correlation coefficient between data in each dimension, if the correlation coefficient between data in two certain dimensions is relatively high, it indicates a strong correlation between them, and they can be considered as a whole when generating feature screening indicators.

[0039] Dynamically select the corresponding dimensionality reduction algorithm according to the feature screening indicators. For sparse high-dimensional data, use the principal component dimensionality reduction algorithm. The principle of the principal component dimensionality reduction algorithm is to transform the original high-dimensional data into a new set of uncorrelated low-dimensional data through a linear transformation, and these new data are called principal components. Assume the original data is , which is a matrix, where is the number of samples, For the feature dimension, new data is obtained after principal component dimensionality reduction. , , is a matrix ( ), is the dimension after dimensionality reduction. For dense low-dimensional data, a factorization algorithm is adopted. The factorization algorithm can factorize the data matrix into the product of multiple low-rank matrices, thereby achieving the purpose of dimensionality reduction. By dynamically selecting the dimensionality reduction algorithm according to the data characteristics in this way, the data can be processed more effectively, and the efficiency and accuracy of feature extraction can be improved.

[0040] After extracting the key feature vectors, data integrity verification is performed on the characteristic parameters. Data integrity verification is achieved by calculating the data missing rate. The calculation formula for the data missing rate is: Data missing rate = (Number of missing data / Total number of data) × 100%. If the verification finds that the data missing rate exceeds the preset second threshold (assuming the second threshold is 10%), the AI modeling module is triggered to interpolate and complete the missing data. When interpolating and completing, the high-priority missing data is the experimental parameter segment that affects the core characteristics. For example, when studying the reactivity of materials, parameters such as reaction temperature and reactant concentration belong to the experimental parameter segment that affects the core characteristics. If these data are missing, they will be preferentially completed. The completion method can adopt algorithms such as linear interpolation and polynomial interpolation, and select the appropriate interpolation method according to the characteristics and distribution of the data.

[0041] Example 4: When the AI modeling module performs modeling, it first constructs a dynamic weight model according to the experimental correlation network of chemical materials. The experimental correlation network describes the mutual relationship between each experimental node in the experimental process, and each experimental node corresponds to a conditional weight coefficient. The conditional weight coefficient reflects the influence degree of this experimental node on the final material characteristics. For example, in the heat treatment experiment of materials, experimental nodes such as heating temperature, holding time, and cooling rate will all affect the final performance of the materials. According to past experimental experience and data analysis, corresponding conditional weight coefficients are assigned to each experimental node.

[0042] Calculate the modeling cooperation weight based on the characteristic differences between adjacent experimental nodes. The characteristic differences can be measured by calculating the difference or distance between the data of two adjacent experimental nodes. Assume that the data of adjacent experimental nodes and are respectively and , and the Euclidean distance can be used to represent their characteristic differences. Calculate the modeling cooperation weight according to the characteristic differences. The modeling cooperation weight is used to adjust the interaction between different experimental nodes, so that the model can better reflect the complex relationships in the experimental process.

[0043] Combined with the experimental historical trend of characteristic parameters, multi-dimensional prediction modeling is carried out for unmeasured experimental nodes. The experimental historical trend is obtained through the analysis of past experimental data, which reflects the law of the change of material characteristics with experimental conditions. For example, by analyzing past experimental data, it is found that the strength of the material first increases and then decreases with the increase of temperature. Using this historical trend, combined with the data of the currently measured experimental nodes and the condition weight coefficient and modeling collaboration weight calculated previously, the data of the unmeasured experimental nodes are predicted. Prediction methods can adopt algorithms such as linear regression and neural networks. Through multi-dimensional calculation and analysis, the predicted values of the unmeasured experimental nodes are obtained, thus completing the multi-dimensional prediction modeling of chemical material characteristics.

[0044] After the modeling is completed, logical verification is carried out on the prediction model. The verification methods include comparing the deviation between the predicted characteristics and the actual measurement of the experimental nodes. The calculation formula for the measurement deviation is: measurement deviation = |predicted value - actual measured value|. If the deviation exceeds the preset third threshold (assuming the third threshold is 5%), the modeling collaboration weight is readjusted and the modeling is iterated. During the iterative modeling process, the parameters of the model are continuously optimized to continuously improve the accuracy of the prediction model until the deviation is less than the third threshold.

[0045] Example 5: When classifying and integrating the prediction model and real-time characteristic parameters using a preset data comparison algorithm, first, the first-level classification labels are divided according to the experimental type. The first-level classification labels include basic physical property types, reaction kinetics types, and structural stability types. The basic physical property types cover experimental data related to basic physical properties such as the density, hardness, and conductivity of the material; the reaction kinetics types mainly include data related to chemical reaction kinetics such as reaction rate and reaction activation energy; the structural stability types involve data such as the crystal structure stability and molecular structure stability of the material.

[0046] Under each first-level classification label, the second-level classification sub-labels are further divided based on the experimental importance. For example, in the basic physical property types, according to the importance and research value of the physical properties, it can be divided into second-level classification sub-labels such as key physical property indicators and general physical property indicators; in the reaction kinetics types, the second-level classification sub-labels can be divided according to the reaction type, such as oxidation reaction, reduction reaction, etc.

[0047] Store the classified experimental data in different partitions of the distributed database according to the label hierarchy. The distributed database can improve the reliability and scalability of data storage. For example, using the Hadoop Distributed File System (HDFS), store the data of key physical property indicators under the basic physical properties category in one partition, and the data of general physical property indicators in another partition; the data of reaction kinetics and structural stability categories are also stored in different partitions according to the corresponding secondary classification sub-labels. In this way, when querying and managing data, the required data can be obtained more conveniently and efficiently.

[0048] To ensure data security and reasonable use, configure the access levels of classification labels according to the preset experimental permissions. For example, for the key physical property indicator data involving core technologies and business secrets, set a higher access level, and only authorized personnel can access it; for some public and general experimental data, a lower access level can be set. When receiving a data query request, verify whether the permission identifier provided by the requester matches the access level of the target classification label. The permission identifier can be the user's identity information, authorization code, etc. If they match, open the data query channel corresponding to the classification label and allow the requester to query the corresponding data; if they do not match, reject the query request to protect the security and privacy of the data.

[0049] Example 6: An AI modeling and data comparison electronic device for analyzing the characteristics of chemical materials, which mainly consists of a multi-source sensing module, a feature extraction module, an AI modeling module, and a data comparison module.

[0050] The multi-source sensing module integrates a variety of sensing devices, which are selected and configured according to the requirements of chemical material experiments. For example, when analyzing metal materials, a spectrometer will be equipped to detect the composition, a scanning electron microscope will be used to obtain physical morphology information, a differential scanning calorimeter will be used to measure thermodynamic properties, and an electrochemical workstation will be used to monitor reaction activity indicators. The multi-source sensing module automatically obtains real-time data from each sensing device according to the preset acquisition frequency and acquisition process, and performs preliminary sorting and preprocessing, and transmits the collected characteristic parameters to the subsequent module.

[0051] After receiving the characteristic parameters transmitted by the multi-source sensing module, the feature extraction module calls the preset feature extraction model. This module will select a suitable processing method according to the characteristics of the characteristic parameters, such as whether they are high-dimensional data. For high-dimensional data, first perform dimensionality reduction processing, and then input it into the feature extraction model. During the operation of the feature extraction model, the extraction threshold is dynamically adjusted according to the characteristics of different material categories in the historical material data, so as to accurately identify and extract the key feature vectors. After the extraction is completed, the feature extraction module outputs the key feature vectors to the AI modeling module.

[0052] After receiving the key feature vectors, the AI modeling module performs multi-dimensional fitting based on the category relevance of chemical materials and the weights of experimental conditions. Functions such as the construction of a dynamic weight model and the calculation of collaborative weights for modeling are implemented inside this module. For example, for a certain type of chemical material, based on its past experimental data and research results, the weights of different experimental conditions are determined, and then combined with the key feature vectors. Through a series of calculations and analyses, a material property prediction model is generated. After generating the model, the AI modeling module outputs the prediction model to the data comparison module.

[0053] The data comparison module obtains the prediction model generated by the AI modeling module and the real-time characteristic parameters collected by the multi-source sensing module, and uses a preset data comparison algorithm for matching verification. This module classifies and integrates the data according to the experimental type and importance, and stores the classified data into the corresponding partitions of the distributed database. At the same time, the data comparison module is also responsible for managing the stored data, verifying the legality of data query requests according to the preset experimental permission configuration, and ensuring the safe and reasonable use of the data.

[0054] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0055] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An AI modeling and data comparison method for analyzing the characteristics of chemical materials, characterized in that, Including: Real-time collection of characteristic parameters of chemical materials through multi-source sensing devices, where the characteristic parameters include composition, physical form, thermodynamic properties, and reaction activity indicators; Input the characteristic parameters into a preset feature extraction model to identify and extract key feature vectors, where the feature extraction model dynamically adjusts the extraction threshold based on the characteristics of different material categories in historical material data; Input the extracted key feature vectors into a preset AI modeling module to generate a material characteristic prediction model, where the AI modeling module performs multi-dimensional fitting according to material category relevance and experimental condition weights; Use a preset data comparison algorithm to match and verify the prediction model and real-time characteristic parameters, and form and store a dynamic analysis data set.

2. The AI modeling and data comparison method for analyzing the characteristics of chemical materials according to claim 1, wherein The steps of constructing the feature extraction model include: Obtain a historical material data set, where each piece of data in the historical material data set is labeled with a material category and a characteristic level; divide training subsets based on the material category and characteristic level, and each training subset corresponds to a material scenario; Parallelly train an initial extraction model using the training subsets until the feature recognition accuracy of the initial extraction model for each material scenario is greater than or equal to a preset first threshold, and then stop training to obtain an intermediate extraction model; Input the historical material data set into the intermediate extraction model, and verify whether the feature recognition results output by the intermediate extraction model meet the preset error range; if they meet, determine the intermediate extraction model as the feature extraction model.

3. The AI modeling and data comparison method for analyzing the characteristics of chemical materials according to claim 1, characterized in that, The real-time collection of characteristic parameters of chemical materials through multi-source sensing devices includes: Establish a communication connection with a target sensing terminal, where the target sensing terminal is deployed at a preset monitoring point of a chemical material experiment device; Continuously read the real-time data of the target sensing terminal at a preset collection frequency, and mark the collection position information based on the spatial distribution characteristics of the real-time data; According to the experimental process topology of the chemical materials, physically align the real-time data at different monitoring points in the same experimental stage to form an associated characteristic parameter set.

4. The AI modeling and data comparison method for analyzing the characteristics of chemical materials according to claim 1, characterized in that Inputting the characteristic parameters into a preset feature extraction model includes: Extract high-dimensional data segments from the characteristic parameters, where the high-dimensional data segments are data segments with parameter dimensions exceeding a preset dimensionality reduction threshold within a continuous collection period; Generate a feature screening index based on the dimensionality distribution and relevance of the high-dimensional data segments; Dynamically select a corresponding dimensionality reduction algorithm according to the feature screening index, where principal component dimensionality reduction algorithm is used for sparse high-dimensional data, and factorization algorithm is used for dense low-dimensional data.

5. The AI modeling and data comparison method for analyzing the characteristics of chemical materials according to claim 4, characterized in that, The method further includes: After extracting the key feature vectors, perform data integrity verification on the characteristic parameters; If it is found through verification that the data missing rate exceeds a preset second threshold, trigger the AI modeling module to interpolate and complete the missing data, where high-priority missing data is an experimental parameter segment that affects core characteristics.

6. The AI modeling and data comparison method for analyzing the characteristics of chemical materials according to claim 1, wherein The AI modeling module includes the following modeling steps: Construct a dynamic weight model according to the experimental association network of the chemical materials, where each experimental node corresponds to a conditional weight coefficient; Calculate and model the collaborative weight based on the characteristic differences of adjacent experimental nodes; Combined with the historical trends of the characteristic parameters, perform multi-dimensional prediction modeling on the unmeasured experimental nodes.

7. The AI modeling and data comparison method for analyzing the characteristics of chemical materials according to claim 6, characterized in that, The method further includes: After the modeling is completed, perform logical verification on the prediction model, where the verification method includes comparing the predicted characteristics with the measurement deviation of the actual experimental nodes; If the deviation exceeds the preset third threshold, readjust the collaborative weight of the modeling and iterate the modeling until the deviation is less than the third threshold.

8. The AI modeling and data comparison method for analyzing the characteristics of chemical materials according to claim 1, wherein Using a preset data comparison algorithm to classify and integrate the prediction model and real-time characteristic parameters includes: Dividing the first-level classification labels according to the experimental type, where the first-level classification labels include basic physical property categories, reaction kinetics categories, and structural stability categories; Under each first-level classification label, further divide the second-level classification sub-labels based on the experimental importance; Store the classified experimental data in different partitions of the distributed database according to the label hierarchy.

9. The AI modeling and data comparison method for analyzing the characteristics of chemical materials according to claim 8, characterized in that The method further includes: Configure the access level of the classification label according to the preset experimental permissions; When receiving a data query request, verify whether the permission identifier provided by the requester matches the access level of the target classification label; If it matches, open the data query channel corresponding to the classification label.

10. An AI modeling and data comparison electronic device for analyzing the characteristics of chemical materials, characterized in that, Includes: A multi-source sensing module for real-time collecting the characteristic parameters of chemical materials through multi-source sensing devices, where the characteristic parameters include composition, physical form, thermodynamic properties, and reaction activity indicators; A feature extraction module for inputting the characteristic parameters into a preset feature extraction model to identify and extract key feature vectors, where the feature extraction model dynamically adjusts the extraction threshold based on the characteristics of different material categories in the historical material data; An AI modeling module for inputting the extracted key feature vectors into a preset AI modeling module to generate a material characteristic prediction model, where the AI modeling module performs multi-dimensional fitting according to the material category relevance and experimental condition weights; A data comparison module for using a preset data comparison algorithm to perform matching verification on the prediction model and real-time characteristic parameters, forming a dynamic analysis data set and storing it.

Images