Heterogeneous industrial data fusion method based on artificial intelligence

Abstract

The invention relates to the technical field of data analysis, and discloses a heterogeneous industrial data fusion method based on artificial intelligence, and the method carries out the structure analysis of a data fusion target, obtains the data structure information, constructs a data fusion target frame, is used for data filling processing, constructs a data fusion channel, and collects element data. According to the method, the key features of the data are extracted, the key features are substituted into the fusion target framework, heterogeneous data fusion is completed, the fusion efficiency of the heterogeneous industrial data is improved, automatic feature extraction and data filling processing are achieved, the intelligence and adaptability of the system are enhanced, and the problem that in the prior art, efficiency is low when large-scale multi-source heterogeneous data is processed is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a method for heterogeneous industrial data fusion based on artificial intelligence. Background Art

[0002] In the industrial field, data comes from various heterogeneous systems, such as different sensors, machines and equipment, management systems, etc. These data usually have different formats, structures and semantics. Traditional data fusion methods mostly rely on manually defined data mapping and conversion rules. This method is incapable of dealing with large-scale data and multi-source heterogeneous data. Traditional data fusion frameworks usually lack flexibility and scalability, and are difficult to adapt to the changing data fusion needs in different industrial scenarios. Summary of the invention

[0003] The purpose of the present invention is to provide a method for heterogeneous industrial data fusion based on artificial intelligence, aiming to solve the problem of low efficiency in processing large-scale multi-source heterogeneous data in the prior art.

[0004] The present invention is implemented in this way. In a first aspect, the present invention provides a method for heterogeneous industrial data fusion based on artificial intelligence, comprising: According to several data fusion elements to be fused with heterogeneous industrial data, a structural analysis of the data fusion target is performed to obtain data structure information of the fusion target; wherein the data fusion element is a data source involved in the fusion of heterogeneous industrial data; Based on the data structure information, a data fusion target framework corresponding to each of the data fusion elements is constructed; wherein the data fusion target framework is used for each of the data fusion elements to perform data filling processing after heterogeneous industrial data fusion; On the basis of the data fusion target framework, a data feature analysis intelligent model is used to construct a corresponding data fusion channel for each of the data fusion elements; wherein the data fusion channel is used to collect data for the data fusion elements; Acquire the element data set from the data fusion element through the data fusion channel, and let the data feature analysis intelligent model extract key data from the element data set to obtain the data key features corresponding to the element data set; Substitute the key features of the data into the data fusion target framework to complete the heterogeneous industrial data fusion of each of the data fusion elements.

[0005] The present invention provides a method for heterogeneous industrial data fusion based on artificial intelligence, which has the following beneficial effects: The present invention performs structural analysis of data fusion targets, obtains data structure information, constructs a data fusion target framework for data filling processing, constructs a data fusion channel, collects element data, extracts key data features, substitutes the key features into the fusion target framework, completes heterogeneous data fusion, improves the fusion efficiency of heterogeneous industrial data, realizes automated feature extraction and data filling processing, enhances the intelligence and adaptability of the system, and solves the problem of low efficiency in processing large-scale multi-source heterogeneous data in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 It is a schematic diagram of the steps of a method for heterogeneous industrial data fusion based on artificial intelligence provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0007] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0008] The implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0009] Reference Figure 1 As shown, a preferred embodiment of the present invention is provided.

[0010] In a first aspect, the present invention provides a method for heterogeneous industrial data fusion based on artificial intelligence, comprising: S1: According to several data fusion elements to be fused with heterogeneous industrial data, a structural analysis of the data fusion target is performed to obtain data structure information of the fusion target; wherein the data fusion element is a data source involved in the fusion of heterogeneous industrial data; S2: Based on the data structure information, construct a data fusion target framework corresponding to each of the data fusion elements; wherein the data fusion target framework is used for each of the data fusion elements to perform data filling processing after heterogeneous industrial data fusion; S3: Based on the data fusion target framework, a data feature analysis intelligent model is used to construct a corresponding data fusion channel for each of the data fusion elements; wherein the data fusion channel is used to collect data for the data fusion elements; S4: acquiring an element data set from the data fusion element through the data fusion channel, and causing the data feature analysis intelligent model to extract key data from the element data set to obtain data key features corresponding to the element data set; S5: Substitute the key features of the data into the data fusion target framework to complete the heterogeneous industrial data fusion of each of the data fusion elements.

[0011] Specifically, in step S1 of the embodiment provided by the present invention, firstly, all data sources that will participate in data fusion are clarified. These data sources include data from different systems, different formats or different structures, such as sensor data, production line control data, ERP system data, etc., so as to determine all elements that will participate in data fusion for subsequent structural analysis.

[0012] More specifically, conduct in-depth structural analysis on each data fusion element and collect its data structure information, including but not limited to data type, field name, field meaning, data relationship (such as primary key, foreign key), constraint conditions (such as non-null, uniqueness), etc., to fully grasp the structural information of each data fusion element and lay the foundation for the structural design of the fusion target.

[0013] More specifically, the structural information of different data elements collected is compared, and the similarities and differences between the various data elements are analyzed, especially the structural differences. It is determined which data fields can be directly matched and which need to be converted or mapped. Possible conflicts (such as repeated field names but different meanings) are identified. The differences between the structures of different data elements are identified and resolved to clear obstacles for the smooth progress of data fusion.

[0014] More specifically, based on the previous analysis, a unified data structure is designed as the data structure of the fusion target. This structure should be able to accommodate the data in all data fusion elements and solve the problem of structural differences, for example, define unified field names and data types, design associations, etc., and build a unified data structure that can cover all data sources as the target structure of the final fusion data.

[0015] More specifically, the designed fusion target structure is verified to ensure that it can effectively accommodate all data elements and be optimized when necessary. The verification methods may include: comparative testing with actual data, simulated data filling, etc., to ensure the rationality, completeness and adaptability of the fusion target structure, so as to provide support for subsequent data fusion.

[0016] It is understandable that through detailed structural analysis, we can ensure a clear understanding of the internal structure of each data fusion element. To avoid data loss, conflict or error caused by unclear structural information during the data fusion process, we design a unified fusion target structure, which can effectively integrate data from different sources. This structure provides a standardized basis for subsequent data filling and fusion, ensuring that the fused data has good operability and consistency. Through structural analysis, possible structural problems can be solved in advance, reducing trial and error and rework in the data fusion process, and improving the efficiency and accuracy of fusion.

[0017] Specifically, in step S2 of the embodiment provided by the present invention, a unified data model is designed based on the data structure information of each data fusion element that has been obtained. The model should include all data fields that need to be fused and resolve differences between data structures, such as the unification of field types, determination of field naming rules, adjustment of data relationships, etc., to create a unified data model to ensure that the data of all data fusion elements can be integrated into this model.

[0018] More specifically, mapping rules should be established for the data fields in each data fusion element and the fields in the unified data model. These rules should clearly define how to map data fields from different sources to the corresponding fields in the target framework, including data conversion, type conversion, unit unification, etc., to ensure that fields from different data sources can be accurately mapped to the unified data model to avoid data loss or conflict.

[0019] More specifically, necessary data integrity constraints are set in the data fusion target framework, including primary key constraints, foreign key constraints, uniqueness constraints, non-null constraints, etc. These constraints ensure that after data fusion, the consistency and integrity of the data are guaranteed. By setting constraints, data inconsistency or violations can be prevented during the data filling process.

[0020] More specifically, according to the needs of data fusion, the initially designed data fusion target framework is optimized, including the elimination of data redundancy, performance improvement (such as index setting), and scalability design of the framework (such as reserving expansion space for future increases in data sources), to ensure that the data fusion target framework can not only meet current data fusion needs, but also adapt to future expansion needs, thereby improving the performance of the overall system.

[0021] More specifically, the constructed data fusion target framework is verified through actual data filling tests to ensure that it can accurately fill data and achieve the expected fusion effect. Simulated data or real data can be used in the verification process to ensure that the data fusion target framework can run stably and reliably in actual applications, laying the foundation for subsequent industrial data fusion.

[0022] It is understandable that by designing a unified data model, heterogeneous industrial data can be integrated into a unified framework to achieve standardized data management, which facilitates subsequent data query, analysis and application. By optimizing the framework, not only the system performance is improved, making data processing faster, but also space is reserved for future expansion needs, making the system have good scalability.

[0023] Specifically, in step S3 of the embodiment provided by the present invention, a data feature analysis intelligent model is used to extract features from the data in each data fusion element. The feature extraction includes but is not limited to data distribution, data type, statistical properties of the data, time series characteristics, spatial characteristics, etc. of the data. The characteristics of each data fusion element are fully understood, and data features that can represent the accurate meaning and key characteristics of the data are extracted from the data fusion elements to achieve normalized statistical processing of fusion element data in various formats.

[0024] More specifically, the data feature analysis intelligent model is trained using an artificial intelligence model to achieve feature extraction of each data fusion element, and the data fusion channel is a data transmission channel built between the data feature analysis intelligent model and each data fusion element, so as to collect the original data corresponding to each data fusion element and allow the data feature analysis intelligent model to perform feature analysis on the original data.

[0025] More specifically, with the support of intelligent models, the construction of data fusion channels has good scalability. Even if the data fusion elements increase or change in the future, the channel system can still adapt to new data collection requirements through model adjustment to ensure the flexibility of the system.

[0026] Specifically, in step S4 of the embodiment provided by the present invention, a complete set of element data is obtained from each data fusion element through the constructed data fusion channel. These data come from different sources, including time series data, spatial data, sensor data, etc. Therefore, these data have different data formats and are obtained based on different data standards. In order to unify these data with different formats and standards, it is necessary to use a data feature analysis intelligent model to extract features from each data, so as to obtain data features with consistent formats and standards.

[0027] More specifically, the data feature analysis intelligent model is initialized, and the parameters and analysis dimensions of the model are set according to the attributes of the feature data set. The model can be a model based on machine learning algorithms, such as cluster analysis, principal component analysis (PCA), factor analysis (FA), neural network, etc., and a suitable analysis model is selected for specific data fusion elements in order to accurately extract key data features.

[0028] More specifically, the initialized data feature analysis intelligent model is used to analyze the feature data set and extract key data features. This step involves reducing the dimensionality of the high-dimensional features of the data and identifying the most representative and influential features in the data. For example, principal components are extracted through PCA, and clustering patterns of data are identified through cluster analysis. Key features that are representative and have decision-making significance are extracted from a large amount of raw data, reducing data redundancy and improving the efficiency and effectiveness of data analysis.

[0029] More specifically, the extracted key data features are ranked by importance to identify the key data that best represents the core features of the element data set. Methods such as feature importance scoring and weight analysis can be used to evaluate the role of each feature in the overall data set.

[0030] More specifically, the extracted key data features are verified to ensure that these features can indeed represent the core information of the feature data set. The verification methods may include cross-validation, model prediction accuracy evaluation, etc. Based on the verification results, the parameters of the intelligent model are optimized to improve the model's analysis and feature extraction capabilities. Through verification and optimization, it is ensured that the extracted key features are highly representative and accurate, so that the model can work stably and efficiently in practical applications.

[0031] It is understandable that through feature analysis and extraction, the original high-dimensional element data set is compressed into more concise and representative key features, which not only reduces the complexity of data storage and calculation, but also improves the efficiency of subsequent data processing. Through the analysis of intelligent models, the extracted key data features can more accurately reflect the core characteristics of data fusion elements. This provides more reliable and meaningful input for subsequent data analysis and decision-making.

[0032] Specifically, in step S5 of the embodiment provided by the present invention, the extracted key features of the data are mapped to the corresponding modules of the data fusion target framework. The mapping process needs to consider the correlation between each feature and the different functional modules in the target framework. Through feature mapping, each key feature is assigned to a suitable fusion logic and processing unit to ensure that the data features can effectively participate in the overall fusion process, achieve seamless connection between the key features and the data fusion target framework, and lay the foundation for subsequent fusion processing.

[0033] More specifically, according to the set rules, the fusion processing of heterogeneous data is performed in the data fusion target framework. The fusion processing may include steps such as merging multi-source data, feature weighting, conflict resolution, and data calibration. The key features of each data fusion element are fused into comprehensive information in this process, providing a basis for the final decision support. Through precise fusion algorithms and processing methods, heterogeneous industrial data from different sources and types are fused into a consistent and usable information set.

[0034] More specifically, the verified fusion data is integrated into the industrial system for further analysis, decision-making and application. The integration process may include data storage, real-time application, report generation, etc. Ensure that the fusion results can effectively support the various functions and business needs of the system, realize the application of data fusion results, provide accurate and reliable data support for the industrial system, and improve the intelligence and automation level of the system.

[0035] It is understandable that the integration problem of heterogeneous industrial data from different data sources and formats is solved through the mapping and fusion of key data features. The fused data has consistency, integrity and availability, and can support more complex analysis and decision-making. The fused data supports higher-level intelligent analysis and automated decision-making, bringing stronger intelligent capabilities to industrial systems. This includes key applications such as real-time monitoring, predictive maintenance, and intelligent scheduling.

[0036] The present invention provides a method for heterogeneous industrial data fusion based on artificial intelligence, which has the following beneficial effects: The present invention performs structural analysis of data fusion targets, obtains data structure information, constructs a data fusion target framework for data filling processing, constructs a data fusion channel, collects element data, extracts key data features, substitutes the key features into the fusion target framework, completes heterogeneous data fusion, improves the fusion efficiency of heterogeneous industrial data, realizes automated feature extraction and data filling processing, enhances the intelligence and adaptability of the system, and solves the problem of low efficiency in processing large-scale multi-source heterogeneous data in the prior art.

[0037] Preferably, the step of performing structural analysis of a data fusion target according to several data fusion elements to be fused with heterogeneous industrial data to obtain data structure information of the fusion target includes: S11: Acquire a historical data set of several data fusion elements to be subjected to heterogeneous industrial data fusion; S12: extracting data features from the historical data set of each data fusion element through a pre-trained data feature extraction intelligent model to obtain element data features of each data fusion element; wherein the element data features are used to describe the data features of the data fed back by the data fusion element; S13: performing commonality analysis on the element data features of each of the data fusion elements to obtain a common format distribution of each of the element data features; wherein the common format distribution includes a plurality of common data features, and the common data features are features of a data format that can be converted from each of the element data features; S14: performing multi-dimensional analysis processing of common conversion on the element data features of each of the data fusion elements based on the common format distribution, so as to obtain common reference features of each of the data fusion elements relative to each of the common data features of the common format distribution; wherein the common reference features include common conversion fitness, common conversion confidence, and common conversion loss; S15: According to the common reference features of each of the data fusion elements relative to each of the common data features of the common format distribution, the best common data feature is selected in the common format distribution as the data structure information of the fusion target.

[0038] Specifically, first, historical data sets are extracted from each element to be subjected to heterogeneous industrial data fusion. These data sets usually include data in multiple formats and structures recorded in the past, which are used to reflect the data representation at different times and in different environments, and provide a comprehensive historical data foundation for subsequent data feature extraction and analysis.

[0039] More specifically, through the pre-trained data feature extraction intelligent model, the historical data set of each data fusion element is analyzed to extract the element data features. These features include the core attributes of the data, such as statistical distribution, trend, periodicity, outliers, etc., which are used to describe the unique characteristics of the data fed back by the element. Representative and descriptive element data features are extracted from the historical data to facilitate subsequent common analysis and conversion.

[0040] More specifically, the commonality analysis is performed on the extracted feature data features to identify and define the common format distribution of these features. The common format distribution refers to the standardized data format into which multiple feature data features can be uniformly converted. The key to this step is to find the commonalities between different feature data features, thereby simplifying the subsequent data fusion process.

[0041] More specifically, based on the common format distribution, the feature data features of each data fusion feature are analyzed in multiple dimensions, and the common reference features relative to the common format distribution are calculated. These common reference features include: common conversion fitness: measuring the degree of adaptability of the conversion of feature data features to common data features; common conversion confidence: indicating the reliability and credibility of the conversion results; common conversion loss: evaluating the information loss that may be generated during the conversion process, and quantifying the degree of matching between each feature data feature and the common data feature through multidimensional analysis, providing a basis for selecting the best common data features.

[0042] More specifically, based on the common reference features of the common format distribution of each element, the common data features that best represent the data structure of the fusion target are evaluated and selected. The selection process considers the features with the highest common conversion fitness, the highest common conversion confidence, and the lowest common conversion loss, to ensure that the final data structure information is both accurate and effective. The best common data features are determined as the final data structure information, providing a unified structural standard for the fusion of heterogeneous industrial data.

[0043] It is understandable that through the establishment of commonality analysis and common format distribution, the standardization of different element data features is achieved, which greatly improves the consistency of data in the data fusion process and avoids data fusion obstacles caused by inconsistent formats. Through multidimensional analysis and processing, the relationship between each element data feature and the common data feature is accurately quantified, making the data fusion process not only accurate but also highly confident, thereby improving the reliability of the fusion results. The application of common format distribution simplifies the data fusion process and no longer needs to deal with complex conversion problems between heterogeneous data, thereby improving the overall fusion efficiency. In the common conversion process, by evaluating the conversion loss, the information loss is minimized, the high fidelity of the data is maintained, and the final data fusion result is ensured to have high-quality information retention.

[0044] Preferably, based on the data structure information, the step of constructing a data fusion target framework corresponding to each of the data fusion elements includes: S21: constructing a corresponding first-layer data framework based on each of the data fusion elements; wherein the first-layer data framework includes a first time coordinate axis and a plurality of element original data axes sequentially arranged on the first time coordinate axis, and the element original data axes are used to store and process the element data sets of each of the data fusion elements; S22: constructing a second-layer data framework based on the data structure information on the first-layer data framework; wherein the second-layer data framework includes a second time coordinate axis and a plurality of element feature data axes sequentially arranged on the second time coordinate axis, the element feature data axes being used to store and process key data features of each of the data fusion elements, and the data structure of the key data features stored in the element feature data axes being consistent with the data structure information; S23: Constructing a third-layer data framework based on the second-layer data framework; wherein the third-layer data framework includes a feature correlation analysis unit and a feature change trend analysis unit, the feature correlation analysis unit is used to analyze the correlation between the key data features of each of the data fusion elements in the second-layer data framework, and the feature change trend analysis unit is used to analyze the change trend of the key data features of each of the data fusion elements in the second-layer data framework.

[0045] Specifically, based on each data fusion element, a first-level data framework is constructed, which includes a first time coordinate axis and a number of element original data axes sequentially arranged on the time coordinate axis. Each element original data axis corresponds to a data fusion element and is used to store the original data set of the element. These data axes are arranged along the time axis, reflecting the evolution of data over time, providing a structured storage mechanism for saving and managing the historical original data of each data fusion element, and providing a basis for subsequent data processing.

[0046] More specifically, based on the first-level data framework, a second-level data framework is constructed using data structure information. The framework includes a second time coordinate axis, and a number of element feature data axes sequentially arranged on the time coordinate axis. Each element feature data axis is used to store key data features of corresponding data fusion elements. The data structure of these features is consistent with the predetermined data structure information. On the basis of the time axis, the key data features of each data fusion element are systematically stored to provide an organized data structure for higher-level analysis.

[0047] More specifically, on top of the second-layer data framework, a third-layer data framework is constructed, which includes two main units: Feature correlation analysis unit: performs correlation analysis on the key data features of each data fusion element in the second-layer data framework, identifies and quantifies the relationship between different elements, which may include linear correlation, nonlinear correlation, time series correlation, etc. Feature change trend analysis unit: analyzes the change trend of the key data features of each data fusion element over time, and helps identify the long-term change pattern or short-term fluctuation law of the feature. Through the correlation and trend analysis of key data features, the internal connection and evolution law between data fusion elements are discovered, providing important reference for further decision support.

[0048] It can be understood that through the construction of the three-layer data framework, a hierarchical structure from raw data to key features to high-level analysis is realized, making the storage, management and analysis of data more systematic and organized. Each layer of the data framework has a clear functional division, which enables the system to perform more efficiently when performing data storage, feature extraction, and association analysis, reducing unnecessary calculations and data redundancy. The correlation and trend analysis unit in the third-layer data framework can handle complex multidimensional data analysis, help identify the deep relationship and dynamic changes between data fusion elements, and provide the system with more comprehensive analysis results. This hierarchical data framework design has strong scalability and can flexibly add new data axes or analysis units according to new data fusion elements or new analysis requirements, thereby continuously enriching and optimizing the data fusion target framework. By storing and managing key data features of a unified structure in the second-layer data framework, data consistency is ensured during the analysis process, thereby improving the accuracy and reliability of data fusion. The analysis unit in the third-layer data framework can generate detailed correlation and trend reports, providing managers and decision-making systems with accurate decision support, and helping to optimize and upgrade industrial systems.

[0049] Preferably, based on the data fusion target framework, the step of using a data feature analysis intelligent model to construct a corresponding data fusion channel for each of the data fusion elements includes: S31: Pre-training the data feature analysis intelligent model to obtain the data feature analysis model that has completed model training; S32: Acquire the data source of each of the data fusion elements, and construct a data collection unit based on the data source of each of the data fusion elements, so as to collect data for the data fusion elements through the data collection unit, thereby acquiring the element data set of the data fusion elements; S33: deploying the data feature analysis intelligent model on the data fusion target framework, so that the data feature analysis intelligent model is data-connected with the data fusion target framework; S34: Connect the data feature analysis intelligent model with each of the data acquisition units so that the data feature analysis intelligent model can extract key data features from the element data set acquired by the data acquisition unit, and transmit the key data features to the corresponding position on the data fusion target framework.

[0050] Specifically, first, pre-training of the data feature analysis intelligent model is carried out. By using large-scale historical data sets and labeled data, the model learns various data feature extraction methods and pattern recognition algorithms so as to accurately extract key data features. After the training is completed, a high-precision data feature analysis model that can be effectively applied to actual data is obtained, ensuring that the data feature analysis intelligent model has high precision and high robustness, and can accurately identify and extract various data features.

[0051] More specifically, obtain the specific data sources of each data fusion element (such as sensors, databases, APIs, etc.), and build corresponding data acquisition units based on these sources. The data acquisition unit is responsible for continuously collecting and processing data from the data source, converting the raw data into a structured element data set, and establishing an effective data acquisition channel for each data fusion element to ensure the real-time and accuracy of data acquisition.

[0052] More specifically, the trained data feature analysis intelligent model is deployed to the data fusion target framework. During the deployment process, it is ensured that the intelligent model can establish a seamless data connection with the data fusion target framework, so that the framework can call the model to analyze and extract data features, and the data fusion target framework has automated data feature analysis capabilities, thereby enhancing the intelligence of the system.

[0053] More specifically, the data acquisition unit is connected to the data feature analysis intelligent model. Through this connection, the model can obtain real-time or batch feature data sets from each data acquisition unit and extract features from these data. The extracted key data features will be transmitted to the corresponding location in the data fusion target framework for storage and further processing, and an automated process between data acquisition, feature extraction and storage will be established to ensure the efficiency and stability of data flow.

[0054] More specifically, the key data features extracted from the intelligent model are stored on the corresponding data axis of the data fusion target framework. These feature data can then be used for advanced data analysis tasks such as correlation analysis and trend prediction, ensuring efficient storage and management of key data features and providing a high-quality data foundation for further decision support and system optimization.

[0055] It can be understood that by introducing the data feature analysis intelligent model, the system can automatically extract key features from the raw data, reduce human intervention, and improve the level of automation in data processing. The efficient connection between the data acquisition unit and the intelligent model ensures real-time collection and processing of data, enabling the system to respond quickly and accurately extract key features to meet the needs of real-time analysis and decision-making.

[0056] Preferably, the steps of pre-training the data feature analysis intelligent model and co-deploying the trained data feature analysis intelligent model with the data fusion target framework include: S311: Prepare a convolutional neural network model, a recurrent neural network model and a training data set; wherein the training data set includes a visual data type set and a sequence data type set; wherein the visual data type set includes visual training data of data fusion elements of a data format corresponding to the visual data type and first target training data of corresponding data key features, and the sequence data type set includes sequence inspection data of data fusion elements of a data format corresponding to the sequence data type and second target training data of corresponding data key features, and the data formats of the first target training data and the second target training data are both consistent with the data structure information; S312: Substituting the visual data type set into the convolutional neural network model, and allowing the convolutional neural network model to perform model training according to the visual data type set, so that the convolutional neural network model can analyze the element data set to obtain data key features corresponding to the data structure information in the data format; S313: Substituting the sequence data type set into the recurrent neural network model, and allowing the recurrent neural network model to perform model training according to the sequence data type set, so that the recurrent neural network model can analyze the element data set to obtain data key features corresponding to the data structure information in the data format; S314: constructing a data type discrimination model based on the convolutional neural network model and the recurrent neural network model, wherein the data type discrimination model is used to discriminate the data format of the data fusion element, and transmit the element data set of the data fusion element to the convolutional neural network model or the recurrent neural network model according to the discrimination result; S315: The convolutional neural network model, the recurrent neural network model and the data type discrimination model that have completed model training are used together as the data feature analysis intelligent model.

[0057] Specifically, first, select and initialize the convolutional neural network model and the recurrent neural network model, and prepare the corresponding training data set. The training data set should include two types of data: a visual data type set: training data containing data fusion elements of the visual data type, which are usually in the form of images, videos or other two-dimensional data, together with the corresponding first target training data, that is, the labeled key data features; a sequence data type set: training data containing data fusion elements of the sequence data type, which are usually in the form of time series, text or other one-dimensional data, together with the corresponding second target training data, that is, the labeled key data features, to ensure that the data required for model training covers different types of features and that the format of these data is consistent with the data structure information required in the final application.

[0058] More specifically, a set of visual data types is substituted into the convolutional neural network model for training. Through multiple iterative optimizations, the model will learn how to extract key features corresponding to the data structure information from the visual data. Ultimately, the trained convolutional neural network model can automatically extract the corresponding key features when a new visual data type is input, thereby obtaining a convolutional neural network model that can efficiently process visual data types, ensuring that it can accurately extract key features from visual data.

[0059] More specifically, a set of sequence data types is substituted into the recurrent neural network model for training. Through iterative training, the model learns how to extract key features corresponding to data structure information from sequence data. Ultimately, the trained recurrent neural network model can automatically extract corresponding key features when new sequence data is input, thereby obtaining a recurrent neural network model that can efficiently process sequence data types, ensuring that it can accurately extract key features from sequence data.

[0060] More specifically, based on the convolutional neural network model and the recurrent neural network model, a data type discrimination model is constructed. The model is used to determine the data format (visual data or sequence data) of the input data fusion elements, and according to the discrimination results, the element data set is transmitted to the corresponding model for processing, ensuring that the system can automatically select the appropriate model according to the data type and optimize the accuracy and efficiency of data feature extraction.

[0061] More specifically, the trained convolutional neural network model, recurrent neural network model, and data type discrimination model are used together as intelligent models for data feature analysis and deployed in the data fusion target framework. These models will automatically process the input feature data set as a whole, extract key data features, and transmit them to the corresponding positions in the data fusion target framework. By integrating multiple models and deploying them together with the data fusion target framework, intelligent and automated data feature extraction and fusion can be achieved.

[0062] It is understandable that through models specially trained for different data types (visual data and sequence data), the system can extract key features more accurately and reduce errors caused by data type mismatch. The convolutional neural network model and the recurrent neural network model are trained for different types of data respectively, which ensures the generalization ability of the model and can adapt to inputs in various data formats. The data type discrimination model automatically determines the data format and selects the appropriate model for processing, avoiding unnecessary waste of computing resources and improving the overall computing efficiency of the system. By deploying the trained model together with the data fusion target framework, the system can automatically complete the entire process from data collection to feature extraction to data fusion, reducing the need for human intervention.

[0063] Preferably, the step of acquiring the element data set from the data fusion element through the data fusion channel and causing the data feature analysis intelligent model to extract key data from the element data set to obtain the data key features corresponding to the element data set includes: S41: Acquire a set of element data from the data fusion element through the data fusion channel; S42: performing data format determination processing on the element data set according to the data type determination model of the data feature analysis intelligent model, and assigning the element data set to the convolutional neural network model or the recurrent neural network model for processing according to the result of the determination processing; S43: Extract key data from the element data set through the convolutional neural network model or the recurrent neural network model to obtain key data features corresponding to the element data set.

[0064] Specifically, through the pre-built data fusion channel, feature data sets from different data fusion elements are acquired in real time or in batches. These data may include visual data (such as images, videos), sequence data (such as time series, text), etc., covering various data formats, ensuring the timeliness and completeness of data collection, and providing a sufficient data foundation for subsequent data processing and feature extraction.

[0065] More specifically, the data type discrimination model in the data feature analysis intelligent model is used to determine the data format of the acquired feature data set. The data type discrimination model will automatically determine whether the data belongs to the visual data type or the sequence data type based on the structure, dimension and content of the data, and assign the data set to the corresponding neural network model (convolutional neural network or recurrent neural network) for processing. Through automated data format discrimination, it ensures that the data set can be sent to the most appropriate model for feature extraction, optimizing the accuracy and efficiency of the processing flow.

[0066] More specifically, according to the data format discrimination results, the feature data sets are sent to the convolutional neural network model or the recurrent neural network model for processing: the visual data type set is sent to the convolutional neural network model, which will use the convolution operation to extract the key features in the visual data such as images and videos; the sequence data type set is sent to the recurrent neural network model, which recursively analyzes the time series, text and other data to extract the key features corresponding to the data structure information, and conducts in-depth analysis of the data set through a specific model to extract valuable key data features for subsequent fusion and decision support.

[0067] More specifically, the key features extracted from the convolutional neural network model or the recurrent neural network model are summarized to form complete data key features corresponding to the element data set. These key features will be transmitted to the corresponding positions of the data fusion target framework for further data fusion, analysis and decision support, generating accurate key data features and ensuring that the system can use these features to achieve smarter analysis and decision-making during the data fusion process.

[0068] It can be understood that through the automated processing of the data type discrimination model, the processes of data format judgment and feature extraction are seamlessly connected, reducing manual intervention and improving the automation level of the system. The convolutional neural network model and the recurrent neural network model respectively extract features for different types of data, ensuring the efficiency and accuracy of processing, and can quickly generate valuable key data features. The system can automatically identify different types of data and select the best processing path, enhancing the intelligence level of the data fusion process, especially when processing multi-source heterogeneous data, enabling more efficient fusion and analysis. Through deep feature extraction for different data types, the system can not only obtain more comprehensive data information, but also deeply explore the implicit features in the data, providing a solid foundation for subsequent advanced data analysis and decision support. This multi-model collaborative working mode enables the system to flexibly respond to changes in different data types, adapt to a variety of application scenarios and business needs, and improve the versatility and scalability of the system.

[0069] Preferably, the step of substituting the key features of the data into the data fusion target framework to complete the heterogeneous industrial data fusion of each of the data fusion elements includes: S51: Substituting the data key features into the second-layer data framework of the data fusion target framework, and allowing the second-layer data framework to store and process the data key features of each of the data fusion elements; S52: Order the third-layer data framework of the data fusion target framework to perform in-depth analysis and processing on the second-layer data framework to obtain the data association features analyzed by the feature association analysis unit, obtain the change trend features analyzed by the feature change trend analysis unit, and store and process the data association features and the change trend features by the third-layer data framework.

[0070] Specifically, the key features of the data that have been extracted through the convolutional neural network and the recurrent neural network models are imported into the second-layer data framework of the data fusion target framework. The function of the second-layer data framework is to store and manage the key features of the data of each data fusion element (such as different types of industrial equipment or sensor data), ensuring that the key features of each industrial data fusion element are systematically stored so that subsequent processing can quickly and accurately access and utilize these features.

[0071] More specifically, in the second-layer data framework, the key features of all data received from different sources are stored and processed. During the storage and processing process, the data format is ensured to be unified and the structure is clear so that subsequent processing and analysis modules can be smoothly connected. A reliable data storage mechanism is established to ensure that the features of multi-source heterogeneous data can be managed in a unified manner and reduce the risk of inconsistent data storage or loss.

[0072] More specifically, in the third-layer data framework, the key features stored in the second-layer data framework are deeply analyzed. The third-layer data framework mainly includes two analysis units: Feature correlation analysis unit: Analyze the correlation features between different data fusion elements. By mining the relationship between key features of the data, identify the potential associations between various elements, such as the collaborative relationship between devices or the mutual influence of sensor data; Feature change trend analysis unit: Analyze the change trend of data features, identify the trend of key features changing over time or operating conditions, such as the gradual decline of equipment performance or the significant fluctuation of certain variables in the production process. Through in-depth analysis, extract the correlation features and change trend features between different industrial data, and provide data basis for subsequent decision support and predictive analysis.

[0073] More specifically, the data association features and change trend features analyzed in the third-layer data framework are stored and processed. These features will serve as the basis for further analysis and decision support, ensuring the continuity and integrity of the data and ensuring that the data analysis results are systematically saved so that the data can be quickly accessed and used to support decision-making when needed.

[0074] It is understandable that by importing the key features of multi-source heterogeneous industrial data into a unified data framework, efficient integration of industrial data is achieved. This integration can establish connections between multiple different data sources, ensuring the interoperability and consistency of data in the entire industrial system. The second-layer data framework provides a unified data storage mechanism to ensure that multi-source data is reliably stored and managed in the system, reducing the risk of data loss or inconsistent formats, and ensuring data integrity. Through the deep analysis function of the third-layer data framework, the system can extract the correlation characteristics and change trends between data. The mining of these features not only helps to understand the inherent relationship of the data, but can also be used to predict potential problems in the system or optimize industrial processes.

[0075] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for heterogeneous industrial data fusion based on artificial intelligence, characterized in that: include: According to several data fusion elements to be fused with heterogeneous industrial data, a structural analysis of the data fusion target is performed to obtain data structure information of the fusion target; wherein the data fusion element is a data source involved in the fusion of heterogeneous industrial data; Based on the data structure information, a data fusion target framework corresponding to each of the data fusion elements is constructed; wherein the data fusion target framework is used for each of the data fusion elements to perform data filling processing after heterogeneous industrial data fusion; On the basis of the data fusion target framework, a data feature analysis intelligent model is used to construct a corresponding data fusion channel for each of the data fusion elements; wherein the data fusion channel is used to collect data for the data fusion elements; Acquire the element data set from the data fusion element through the data fusion channel, and let the data feature analysis intelligent model extract key data from the element data set to obtain the data key features corresponding to the element data set; Substitute the key features of the data into the data fusion target framework to complete the heterogeneous industrial data fusion of each of the data fusion elements.

2. The method for heterogeneous industrial data fusion based on artificial intelligence according to claim 1, characterized in that: According to several data fusion elements to be fused with heterogeneous industrial data, the steps of performing structural analysis on the data fusion target to obtain data structure information of the fusion target include: Acquire a historical data set of several data fusion elements to be subjected to heterogeneous industrial data fusion; Through the pre-trained data feature extraction intelligent model, the historical data set of each data fusion element is subjected to data feature extraction processing to obtain the element data feature of each data fusion element; wherein the element data feature is used to describe the data feature of the data fed back by the data fusion element; Performing commonality analysis on the element data features of each of the data fusion elements to obtain a common format distribution of each of the element data features; wherein the common format distribution includes a plurality of common data features, and the common data features are features of a data format that can be converted from each of the element data features; Based on the common format distribution, a multi-dimensional analysis process of common conversion is performed on the element data features of each of the data fusion elements to obtain common reference features of each of the data fusion elements relative to each of the common data features of the common format distribution; wherein the common reference features include common conversion fitness, common conversion confidence, and common conversion loss; According to the common reference features of each of the data fusion elements relative to each of the common data features of the common format distribution, the best common data feature is selected in the common format distribution to serve as the data structure information of the fusion target.

3. The method for heterogeneous industrial data fusion based on artificial intelligence according to claim 1, characterized in that: Based on the data structure information, the step of constructing a data fusion target framework corresponding to each of the data fusion elements includes: Constructing a corresponding first-layer data framework based on each of the data fusion elements; wherein the first-layer data framework includes a first time coordinate axis and a plurality of element original data axes sequentially arranged on the first time coordinate axis, and the element original data axes are used to store and process the element data sets of each of the data fusion elements; A second layer of data framework is constructed based on the data structure information on the first layer of data framework; wherein the second layer of data framework includes a second time coordinate axis and a plurality of element feature data axes sequentially arranged on the second time coordinate axis, the element feature data axes are used to store and process key data features of each of the data fusion elements, and the data structure of the key data features stored in the element feature data axes is consistent with the data structure information; A third-layer data framework is constructed on the basis of the second-layer data framework; wherein the third-layer data framework includes a feature correlation analysis unit and a feature change trend analysis unit, the feature correlation analysis unit is used to analyze the correlation between the key data features of each of the data fusion elements in the second-layer data framework, and the feature change trend analysis unit is used to analyze the change trend of the key data features of each of the data fusion elements in the second-layer data framework.

4. The method for heterogeneous industrial data fusion based on artificial intelligence according to claim 1, characterized in that: Based on the data fusion target framework, the steps of using the data feature analysis intelligent model to construct a corresponding data fusion channel for each of the data fusion elements include: Pre-training the data feature analysis intelligent model to obtain the data feature analysis model that has completed model training; Acquire the data source of each of the data fusion elements, and construct a data collection unit based on the data source of each of the data fusion elements, so as to collect data for the data fusion elements through the data collection unit, thereby acquiring the element data set of the data fusion elements; Deploy the data feature analysis intelligent model on the data fusion target framework, and connect the data feature analysis intelligent model with the data fusion target framework; The data feature analysis intelligent model is data-connected to each of the data acquisition units so that the data feature analysis intelligent model can extract key data features from the element data set acquired by the data acquisition unit and transmit the key data features to the corresponding position on the data fusion target framework.

5. The method for heterogeneous industrial data fusion based on artificial intelligence as claimed in claim 4 is characterized in that: The steps of pre-training the data feature analysis intelligent model and co-deploying the trained data feature analysis intelligent model with the data fusion target framework include: Prepare a convolutional neural network model, a recurrent neural network model and a training data set; wherein the training data set includes a visual data type set and a sequence data type set; wherein the visual data type set includes visual training data of data fusion elements of the visual data type whose data format corresponds to the first target training data of the corresponding data key features, and the sequence data type set includes sequence inspection data of data fusion elements of the sequence data type whose data format corresponds to the second target training data of the corresponding data key features, and the data formats of the first target training data and the second target training data are consistent with the data structure information; Substituting the visual data type set into the convolutional neural network model, allowing the convolutional neural network model to perform model training according to the visual data type set, so that the convolutional neural network model can analyze the element data set to obtain data key features corresponding to the data structure information in the data format; Substituting the sequence data type set into the recurrent neural network model, allowing the recurrent neural network model to perform model training according to the sequence data type set, so that the recurrent neural network model can analyze the element data set to obtain data key features corresponding to the data structure information in a data format; A data type discrimination model is constructed based on the convolutional neural network model and the recurrent neural network model, wherein the data type discrimination model is used to discriminate the data format of the data fusion element, and transmit the element data set of the data fusion element to the convolutional neural network model or the recurrent neural network model according to the discrimination result; The convolutional neural network model, the recurrent neural network model and the data type discrimination model that have completed model training are collectively used as the data feature analysis intelligent model.

6. The method for heterogeneous industrial data fusion based on artificial intelligence according to claim 5, characterized in that: The steps of obtaining a set of element data from the data fusion element through the data fusion channel and causing the data feature analysis intelligent model to extract key data from the set of element data to obtain key features of data corresponding to the set of element data include: Acquire a set of element data from the data fusion element through the data fusion channel; Performing data format determination processing on the element data set according to the data type determination model of the data feature analysis intelligent model, and assigning the element data set to the convolutional neural network model or the recurrent neural network model for processing according to the result of the determination processing; The key data of the element data set is extracted through the convolutional neural network model or the recurrent neural network model to obtain key data features corresponding to the element data set.

7. The method for heterogeneous industrial data fusion based on artificial intelligence according to claim 1, characterized in that: Substituting the key features of the data into the data fusion target framework to complete the heterogeneous industrial data fusion of each of the data fusion elements includes: Substituting the data key features into the second-layer data framework of the data fusion target framework, and allowing the second-layer data framework to store and process the data key features of each of the data fusion elements; The third-layer data framework of the data fusion target framework performs in-depth analysis and processing on the second-layer data framework to obtain the data association features analyzed by the feature association analysis unit, and obtain the change trend features analyzed by the feature change trend analysis unit, and the third-layer data framework stores and processes the data association features and the change trend features.

Images