Model construction method based on data driving in industrial field
By obtaining device data in industrial Internet of Things scenarios and building resource models in combination with industry specifications and expert opinions, the problem of time-consuming and difficult optimization of industrial models is solved, and high accuracy and flexibility of equipment prediction and management is achieved.
Patent Information
- Application Number
- CN202411619864.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-13
- Publication Date
- 2025-02-25
AI Technical Summary
In the industrial Internet of Things scenario, the industrial model construction process takes a long time, is difficult to optimize, and is complex to migrate.
By obtaining real-time and historical data of the equipment, building a resource model in combination with industry field specifications and business expert guidance, generating the basic model of the equipment and predicting the operating data of the equipment. The specific steps include configuring the sensor to obtain real-time data, building a resource library, building a built-in and AI models to jointly generate basic models, iteratively train and adjusting model indicators.
Improves the accuracy and flexibility of equipment prediction, reduces manual intervention, and realizes intelligent management and data-driven decision-making.
Smart Images
Figure CN119514363A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of mechanism models, and in particular to a data-driven model building method in the industrial field. Background Art
[0002] At present, in the field of industrial Internet of Things, data collection solutions and industrial model construction solutions based on specific scenarios have already had extensive technical support and application cases. However, in the industrial Internet of Things scenario, the process of industrial model construction usually involves problems such as long time consumption, difficult optimization, and complex migration.
[0003] Therefore, the present invention proposes a data-driven model building method in the industrial field. Summary of the invention
[0004] The present invention provides a data-driven model building method in the industrial field, which is used to generate a basic model of the equipment and predict the operating data of the equipment by acquiring real-time and historical data of the equipment and combining it with a resource model built according to industry standards and guidance from business experts.
[0005] In one aspect, the present invention provides a data-driven model building method in an industrial field, comprising: Step 1: configure the sensor to be installed in the target device, obtain the real-time data of the target device, and obtain the historical data of the target device according to the historical database; Step 2: Build a resource library based on industry standards and expert database information; Step 3: Build a built-in model based on the built-in algorithm of the parameters of the target device, call a third-party AI service to build an AI model, and the basic model is generated by the built-in model and the AI model; Step 4: Generate a training set based on the historical data of the target device and the resource instance obtained by matching the target device to the resource library, and use the training set to train the basic model to iteratively generate the first model; Step 5: Use the real-time data of the target device as a test set, substitute the test set data into the first model to generate a prediction result, compare the prediction result with the actual result of the target device, and adjust the model indicators of the first model to obtain the final model.
[0006] On the other hand, configuring the sensor to be installed in the target device to obtain real-time data of the target device includes: According to the functional requirements of the target device, match the sensor type corresponding to the functional requirements; Determining an installation position of the sensor on the target device according to installation requirements of any of the sensors; A unique first number is configured for the sensor, a unique second number is configured for the installation position, and matching of the sensor and the installation position is completed according to a one-to-one correspondence between the sensor and the installation position; Complete the configuration and startup at the corresponding installation location according to the installation instructions of the sensor, and obtain the real-time data of the corresponding sensor in real time.
[0007] On the other hand, obtaining the historical data of the target device according to the historical database includes: Connect to the historical database according to the preset basic database connection information: Select one or more parameter types as the filtering conditions, and obtain the filtering statement conditions based on the parameter type-parameter code mapping table; A filter statement is generated according to the filter statement conditions and imported into a historical database to obtain historical data corresponding to the parameter type.
[0008] On the other hand, it also includes: The historical data and real-time data of any parameter type are processed into a unified unit based on a preset standard unit to obtain first real-time data and first historical data; Based on the standard time interval, the first real-time data and the first historical data are time aligned to align the starting points, and then the time of all the first real-time data and the first historical data is time standardized according to the standard time interval, and the second data point corresponding to the standardized time node is processed according to the alignment function to obtain standardized real-time data and standardized historical data of the parameter type.
[0009] On the other hand, in accordance with industry standards and combined with expert database information, a resource library is built, including: Obtaining industry specifications and standards for the target device and determining a basic framework of a resource model for the target device; According to the specific professional application information of the target device in the target device matching expert database, the business parameters of the resource model basic framework are adjusted, and the resource model is constructed; Build a resource library and connect to the business data source to obtain original resource data, generate standard resource data after processing through the resource model, and store the resource data of the target device in the resource library.
[0010] On the other hand, a built-in model is constructed according to the built-in algorithm of the parameters of the target device, and a third-party AI service is called to construct an AI model. The basic model is jointly generated by the built-in model and the AI model, including: Obtain all process parameters of the target device, and select necessary process parameters of the target device based on the functional requirements of the target device: ; Indicates the necessity of any process parameter. represents the standard deviation of the mean of the process parameter and the standard process threshold, represents the standardized historical data of the process parameter at the i-th time node, represents the standard data of the i-th time node of the process parameter, ( )express and dot product, represents the variance of the standardized historical data of the process parameter, n represents that there are n standardized historical data of the process parameter in total, represents the maximum value of the standardized historical data of the process parameter, represents the minimum value of the standardized historical data of the process parameter; Screening the degree of necessity of process parameters according to a minimum degree of necessity screening threshold, and the degree of necessity greater than the minimum degree of necessity is the necessary process parameter of the target device; Selecting a built-in statistical formula according to the necessary process parameters of the target device, verifying the validity of the built-in statistical formula and adjusting the configuration coefficients of the formula parameters of the necessary process parameters to complete the configuration of the built-in model; Design the interface of third-party AI services and allow custom request parameters. Use the request parameters to call the AI model to obtain AI calculation results. The basic model of the target device is jointly constructed by the AI model and the built-in model, supports real-time comparison of AI calculation results and built-in model calculation results, and supports free switching between the two models.
[0011] On the other hand, according to the historical data of the target device, a training set is generated in combination with a resource instance obtained by matching the target device to a resource library, and a basic model is trained using the training set to iteratively generate a first model, including: Acquire standardized historical data of the target device and extract features of the historical data of any parameter type; Match resource instances whose feature matching distance with historical data of parameter types is greater than the clustering standard from the resource library, and generate resource instance clusters, and use the resource instance clusters as training set units of corresponding parameter types; The training set units of all parameter types are obtained to generate a training set, and the basic model is iteratively trained based on the training set to generate a first model.
[0012] On the other hand, the real-time data of the target device is used as a test set, the test set data is substituted into the first model to generate a prediction result, the prediction result is compared with the actual result of the target device, and the model index of the first model is adjusted to obtain a final model, including: Acquire standardized real-time data of the target device, use the standardized real-time data as a test set, and substitute the test set into the first model to generate a prediction result; The prediction results are split according to the parameter type to obtain the prediction parameter results, and the prediction parameter results are compared with the actual results of the corresponding parameter types of the real-time data to obtain the deviation value; If the deviation value is greater than the adjustment threshold, the model indicators of the first model are adjusted according to the deviation value to obtain the final model indicators, and the first model is retrained using the final model indicators to generate the final model. If the deviation value is less than the adjustment threshold, the first model is set as the final model.
[0013] On the other hand, it includes: The final model and its parameters are documented and ready to be deployed in a real-world environment for real-time predictions.
[0014] On the other hand, it includes: The first model retraining triggers the retraining mechanism by detecting whether the final model indicators have changed.
[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a data-driven model building method in the industrial field, which is used to generate a basic model of the equipment and predict the operating data of the equipment by acquiring real-time and historical data of the equipment and combining it with a resource model built according to industry standards and guidance from business experts. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0017] Figure 1 It is a flowchart of a data-driven model building method in an industrial field provided by an embodiment of the present invention; Figure 2 This is a diagram illustrating a trigger. DETAILED DESCRIPTION
[0018] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0019] Embodiment 1: like Figure 1 As shown, an embodiment of the present invention provides a data-driven model building method in an industrial field, comprising: Step 1: configure the sensor to be installed in the target device, obtain the real-time data of the target device, and obtain the historical data of the target device according to the historical database; Step 2: Build a resource library based on industry standards and expert database information; Step 3: Build a built-in model based on the built-in algorithm of the parameters of the target device, call a third-party AI service to build an AI model, and the basic model is generated by the built-in model and the AI model; Step 4: Generate a training set based on the historical data of the target device and the resource instance obtained by matching the target device to the resource library, and use the training set to train the basic model to iteratively generate the first model; Step 5: Use the real-time data of the target device as a test set, substitute the test set data into the first model to generate a prediction result, compare the prediction result with the actual result of the target device, and adjust the model indicators of the first model to obtain the final model.
[0020] In this embodiment, the sensor is a device used to detect and measure physical quantities (such as temperature, humidity, pressure, etc.) and convert these data into signals that can be analyzed and processed.
[0021] In this embodiment, real-time data refers to current data obtained from a sensor or device at a specific time point, reflecting the instantaneous status of the target device.
[0022] In this embodiment, the history library is a database that stores data collected in the past, including operating status, performance indicators and other relevant information about the target device at different time points.
[0023] In this embodiment, historical data refers to relevant data about the target device collected within a certain period of time in the past.
[0024] In this embodiment, the industry specification standards are guiding principles and technical requirements developed for specific industries, formulated by industry associations, government agencies or standardization organizations, and cover all aspects of design, manufacturing, testing and operation.
[0025] In this embodiment, the expert database information refers to relevant data of experts with rich experience and knowledge in a specific field, including: experts' research results, industry best practices, technical suggestions, case analysis and solutions, etc.
[0026] In this embodiment, the resource library is a database that integrates a variety of information and tools, including standards, technical documents, best practices, case studies, expert knowledge, etc. related to the target device.
[0027] In this embodiment, the parameter built-in algorithm refers to an algorithm preset in a device or system. For example, in the case of sensor monitoring, these algorithms may involve functions such as data filtering, anomaly detection, and trend analysis.
[0028] In this embodiment, the built-in model refers to a model that is integrated in the target device and built based on specific parameters and algorithms.
[0029] In this embodiment, AI services refer to various artificial intelligence functions and tools provided through cloud computing or local platforms, including pre-trained models, algorithm interfaces, and data processing services.
[0030] In this embodiment, the AI model is a mathematical model trained by machine learning or deep learning algorithms, and is used to recognize patterns, make predictions, or automatically perform tasks.
[0031] In this embodiment, the basic model is a preliminary model generated by the built-in model and the AI model.
[0032] In this embodiment, the training set is a data set used to train a machine learning model, and is composed of historical data of the target device and matching resource instances.
[0033] In this embodiment, the first model is a preliminary model generated by iteratively training the basic model through the training set.
[0034] In this embodiment, the test set is a data set used to evaluate model performance, and is composed of real-time data of the target device.
[0035] In this embodiment, the prediction result is the output generated by inputting the test set data into the first model, and represents the model's prediction of the target device state.
[0036] In this embodiment, the model indicators are quantitative standards for evaluating model performance, including accuracy, precision, recall, mean square error, etc.
[0037] In this embodiment, the final model refers to the adjusted first model.
[0038] The working principle and beneficial effects of the above technical solution are: through real-time data collection, resource library construction and model optimization, the accuracy and flexibility of equipment prediction are improved, manual intervention is reduced, intelligent management is realized, data-driven decision-making is supported, and system performance is continuously improved.
[0039] Embodiment 2: On the basis of the above-mentioned embodiment 1, configuring a sensor to be installed in a target device to obtain real-time data of the target device includes: According to the functional requirements of the target device, match the sensor type corresponding to the functional requirements; Determining an installation position of the sensor on the target device according to installation requirements of any of the sensors; A unique first number is configured for the sensor, a unique second number is configured for the installation position, and matching of the sensor and the installation position is completed according to a one-to-one correspondence between the sensor and the installation position; Complete the configuration and startup at the corresponding installation location according to the installation instructions of the sensor, and obtain the real-time data of the corresponding sensor in real time.
[0040] In this embodiment, the functional requirements refer to the specific capabilities and performances that the target device needs to have, including: data collection, response capability, compatibility, etc.
[0041] In this embodiment, the requirement type refers to the classification and characteristics of sensors required according to the functional requirements of the target device, including: environment detection type, state detection type, electrical detection type, etc.
[0042] In this embodiment, the installation requirements refer to the specific conditions and requirements that need to be met when the sensor is installed on the target device, including: space requirements, environmental adaptability, interface compatibility, etc.
[0043] In this embodiment, the first number refers to a unique identifier assigned to each sensor.
[0044] In this embodiment, the second number refers to a unique identifier assigned to each sensor installation location.
[0045] In this embodiment, the corresponding relationship refers to the matching relationship between the sensor and its installation location, that is, the unique pairing between each sensor (identified by the first number) and its specific installation location (identified by the second number).
[0046] The working principle and beneficial effects of the above technical solution are: by matching sensors with equipment requirements, clarifying installation locations and numbering, ensuring accurate configuration and startup, acquiring data in real time, improving monitoring accuracy and system reliability, optimizing resource utilization, and enhancing intelligent management capabilities.
[0047] Embodiment 3: Based on the above embodiment 1, it is characterized in that obtaining the historical data of the target device according to the historical database includes: Connect to the historical database according to the preset basic database connection information: Select one or more parameter types as the filtering conditions, and obtain the filtering statement conditions based on the parameter type-parameter code mapping table; A filter statement is generated according to the filter statement conditions and imported into a historical database to obtain historical data corresponding to the parameter type.
[0048] In this embodiment, the historical database is a database for storing historical data, and is used to save past records and data.
[0049] In this embodiment, the parameter type-parameter code mapping table is a table indicating the mapping relationship between parameter type and parameter code.
[0050] In this embodiment, the filter statement condition is a specific condition obtained from a parameter type-parameter code mapping table according to the parameter type selected by the user.
[0051] The working principle and beneficial effects of the above technical solution are: by connecting to the historical database and generating filtering statements based on parameter types, relevant historical data can be effectively extracted, data access flexibility and efficiency can be improved, intelligent decision-making can be supported, and system responsiveness and accuracy can be enhanced.
[0052] Embodiment 4: Based on the above embodiment 1, it also includes: The historical data and real-time data of any parameter type are processed into a unified unit based on a preset standard unit to obtain first real-time data and first historical data; Based on the standard time interval, the first real-time data and the first historical data are time aligned to align the starting points, and then the time of all the first real-time data and the first historical data is time standardized according to the standard time interval, and the second data point corresponding to the standardized time node is processed according to the alignment function to obtain standardized real-time data and standardized historical data of the parameter type.
[0053] In this embodiment, the preset standard unit is a unified measurement standard used to convert and unify data from different sources (such as historical data and real-time data), including time, length, weight, temperature, etc.
[0054] In this embodiment, the unit normalization process is to convert data from different sources into the same standard unit, for example, converting inches into centimeters, or converting Fahrenheit into Celsius.
[0055] In this embodiment, the first real-time data is real-time data that has been processed by unit unification and has been converted into a preset standard unit.
[0056] In this embodiment, the first historical data is historical data that has been processed by unit unification and has been converted into a preset standard unit.
[0057] In this embodiment, the standard time interval is a preset time unit used to align and standardize time points during data processing.
[0058] In this embodiment, the time alignment operation is a process of matching and synchronizing data from different time sources.
[0059] In this embodiment, time standardization is a process of converting data at different time points into a unified format, which involves resampling or interpolating all aligned data points according to a preset standard time interval.
[0060] In this embodiment, the alignment function is a mathematical algorithm tool used to map the time points of two data sets to the same time nodes during the time alignment process. Interpolation, averaging or other aggregation methods are usually used to process the first real-time data and the first historical data accordingly at the standardized time nodes to generate consistent standardized data.
[0061] In this embodiment, the standardized real-time data is real-time data that conforms to a preset standard format after being time-aligned and unified in units.
[0062] In this embodiment, the standardized historical data is historical data that conforms to a preset standard format after being processed by time alignment and unit unification.
[0063] The working principle and beneficial effects of the above technical solution are: to process historical data and real-time data through unit unification and time alignment, to ensure data consistency and comparability, to generate standardized data, to improve analysis accuracy and decision support capabilities, and to enhance the practicality of the system.
[0064] Embodiment 5: Based on the above embodiment 1, a resource library is constructed in accordance with industry standards and combined with expert database information, including: Obtaining industry specifications and standards for the target device and determining a basic framework of a resource model for the target device; According to the specific professional application information of the target device in the target device matching expert database, the business parameters of the resource model basic framework are adjusted, and the resource model is constructed; Build a resource library and connect to the business data source to obtain original resource data, generate standard resource data after processing through the resource model, and store the resource data of the target device in the resource library.
[0065] In this embodiment, industry specifications and standards refer to rules, guidelines and requirements formulated within a specific industry, including technical standards, safety standards, quality standards, etc.
[0066] In this embodiment, the resource model is a structured description of specific equipment or system resources, including: resource definition, resource relationship, data structure, business parameters and other contents.
[0067] In this embodiment, the basic framework refers to the basic structure and organization method of building a resource model, including: component definition, attributes and relationships, data flow and other contents.
[0068] In this embodiment, the specific professional application information refers to detailed technical and business data related to the target device, including: application scenarios, performance indicators, operating specifications, etc.
[0069] In this embodiment, the business parameter adjustment refers to modifying and optimizing the parameters in the resource model basic framework according to the specific professional application information of the target device.
[0070] In this embodiment, the original resource data refers to the original information collected during the operation of the target device without being processed or analyzed.
[0071] In this embodiment, standard resource data refers to processed and standardized original resource data, which has a format and structure that complies with industry standards and business requirements.
[0072] The working principle and beneficial effects of the above technical solution are: building a target device resource model through industry standards, adjusting business parameters, obtaining original data and generating standard resource data, and finally storing it in a resource library, thereby improving the standardization, availability and analysis value of the data and facilitating subsequent management and application.
[0073] Embodiment 6: On the basis of the above-mentioned embodiment 1, a built-in model is constructed according to the built-in algorithm of the parameters of the target device, and a third-party AI service is called to construct an AI model. The basic model is jointly generated by the built-in model and the AI model, including: Obtain all process parameters of the target device, and select necessary process parameters of the target device based on the functional requirements of the target device: ; Indicates the necessity of any process parameter. represents the standard deviation of the mean of the process parameter and the standard process threshold, represents the standardized historical data of the process parameter at the i-th time node, represents the standard data of the i-th time node of the process parameter, ( )express and dot product, represents the variance of the standardized historical data of the process parameter, n represents that there are n standardized historical data of the process parameter in total, The maximum value of the standardized historical data of the process parameter, represents the minimum value of the standardized historical data of the process parameter; Screening the degree of necessity of process parameters according to a minimum degree of necessity screening threshold, and the degree of necessity greater than the minimum degree of necessity is the necessary process parameter of the target device; Selecting a built-in statistical formula according to the necessary process parameters of the target device, verifying the validity of the built-in statistical formula and adjusting the configuration coefficients of the formula parameters of the necessary process parameters to complete the configuration of the built-in model; Design the interface of third-party AI services and allow custom request parameters. Use the request parameters to call the AI model to obtain AI calculation results. The basic model of the target device is jointly constructed by the AI model and the built-in model, supports real-time comparison of AI calculation results and built-in model calculation results, and supports free switching between the two models.
[0074] In this embodiment, process parameters refer to various variables and indicators used to describe and control a specific process or procedure during the manufacturing or production process, such as temperature, pressure, time, concentration, etc.
[0075] In this embodiment, the necessary process parameters refer to parameters that are critical to achieving functional requirements in a specific device or process.
[0076] In this embodiment, the degree of necessity is an indicator that measures the importance of a specific process parameter to the functional requirements of the target device, indicating the criticality of the parameter in meeting the performance, quality and efficiency of the device.
[0077] In this embodiment, the standard deviation is a parameter indicating the degree of deviation obtained by calculating the Euclidean distance between the mean of the process parameter and the standard process threshold.
[0078] In this embodiment, the minimum necessary screening threshold process parameter represents the minimum matching degree required to achieve the required functional requirements.
[0079] In this embodiment, the built-in statistical formula is usually a mathematical model used to analyze and process process parameters, such as mean calculation, variance calculation, regression analysis, etc.
[0080] In this embodiment, the configuration coefficient refers to the weight ratio used to adjust and optimize parameters in the model or algorithm.
[0081] In this embodiment, the interface is to define the interaction mode between different systems and exchange information based on the interface.
[0082] In this embodiment, request parameters refer to parameters that users can customize and pass to the service when interacting with the third-party AI service interface, such as time, calculation type, process parameter name, etc.
[0083] The working principle and beneficial effects of the above technical solution are: by screening necessary process parameters, configuring built-in statistical models, and designing AI interfaces, a basic model combining AI and built-in models is constructed to achieve real-time comparison and flexible switching, thereby improving equipment performance and intelligence.
[0084] Embodiment 7: On the basis of the above-mentioned embodiment 4, a training set is generated according to the historical data of the target device and the resource instance obtained by matching the target device to the resource library, and a basic model is trained by using the training set to iteratively generate a first model, including: Acquire standardized historical data of the target device and extract features of the historical data of any parameter type; Match resource instances whose feature matching distance with historical data of parameter types is greater than the clustering standard from the resource library, and generate resource instance clusters, and use the resource instance clusters as training set units of corresponding parameter types; The training set units of all parameter types are obtained to generate a training set, and the basic model is iteratively trained based on the training set to generate a first model.
[0085] In this embodiment, the feature is a prominent attribute indicator used to describe the data, such as statistical features, time series features, distribution features, etc.
[0086] In this embodiment, the clustering criterion refers to a criterion for evaluating the degree of aggregation of resource instances in the feature space.
[0087] In this embodiment, a resource instance refers to a data object with specific attributes or characteristics in a specific context, including: a device, a sensor, a software component, etc.
[0088] In this embodiment, the resource instance cluster refers to a set formed by clustering a group of resource instances according to feature similarity.
[0089] In this embodiment, the training set unit refers to a basic data unit used for model training, which is composed of a group of related resource instance clusters.
[0090] The working principle and beneficial effects of the above technical solution are: by extracting parameter features and matching resource instances, forming training set units, iteratively training the basic model, and generating the first model, thereby improving model accuracy and adaptability and optimizing device performance.
[0091] Embodiment 8: On the basis of the above-mentioned embodiment 1, the real-time data of the target device is used as a test set, the test set data is substituted into the first model to generate a prediction result, the prediction result is compared with the actual result of the target device, and the model index of the first model is adjusted to obtain a final model, including: Acquire standardized real-time data of the target device, use the standardized real-time data as a test set, and substitute the test set into the first model to generate a prediction result; The prediction results are split according to the parameter type to obtain the prediction parameter results, and the prediction parameter results are compared with the actual results of the corresponding parameter types of the real-time data to obtain the deviation value; If the deviation value is greater than the adjustment threshold, the model indicators of the first model are adjusted according to the deviation value to obtain the final model indicators, and the first model is retrained using the final model indicators to generate the final model. If the deviation value is less than the adjustment threshold, the first model is set as the final model.
[0092] In this embodiment, the deviation value refers to the difference between the predicted result and the actual result.
[0093] In this embodiment, the adjustment threshold is a preset value used to determine whether the prediction result of the model is within an acceptable range.
[0094] In this embodiment, the final model index is an index used to evaluate the model performance after adjustment and optimization.
[0095] The working principle and beneficial effects of the above technical solution are: by using real-time data as a test set, generating predictions and calculating deviation values, adjusting model indicators based on the deviations, ensuring the accuracy and reliability of the final model, and optimizing prediction performance.
[0096] Embodiment 9: On the basis of the above-mentioned embodiment 8, it includes: The final model and its parameters are documented and ready to be deployed in a real-world environment for real-time predictions.
[0097] In this embodiment, documentation refers to the process of systematically recording and organizing information such as the final model and its parameters, design ideas, usage methods and maintenance guidelines, including writing detailed documents to describe the model's architecture, data sources, algorithm selection, parameter settings, etc.
[0098] The working principle and beneficial effects of the above technical solution are: documenting the final model facilitates deployment and maintenance, and improves the efficiency and reliability of real-time prediction.
[0099] Embodiment 10: On the basis of the above-mentioned embodiment 8, it includes: The first model retraining triggers the retraining mechanism by detecting whether the final model indicators have changed.
[0100] In this embodiment, a trigger is an automated mechanism that monitors the occurrence of a specific condition or event, such as Figure 2 shown.
[0101] The working principle and beneficial effects of the above technical solution are: trigger monitoring indicator changes, automatic retraining to ensure model optimization, and improve prediction accuracy and stability.
[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A data-driven model building method in the industrial field, characterized in that: include: Step 1: configure the sensor to be installed in the target device, obtain the real-time data of the target device, and obtain the historical data of the target device according to the historical database; Step 2: Build a resource library based on industry standards and expert database information; Step 3: Build a built-in model according to the built-in algorithm of the parameters of the target device, call a third-party AI service to build an AI model, and the basic model is generated by the built-in model and the AI model; Step 4: Generate a training set based on the historical data of the target device and the resource instance obtained by matching the target device to the resource library, and use the training set to train the basic model to iteratively generate the first model; Step 5: Use the real-time data of the target device as a test set, substitute the test set data into the first model to generate a prediction result, compare the prediction result with the actual result of the target device, and adjust the model indicators of the first model to obtain the final model.
2. The method for constructing a model based on data-driven in the industrial field according to claim 1, characterized in that: The configuration sensor is installed in the target device to obtain real-time data of the target device, including: According to the functional requirements of the target device, match the sensor type corresponding to the functional requirements; Determine the installation position of the sensor on the target device according to the installation requirements of any of the sensors; A unique first number is configured for the sensor, a unique second number is configured for the installation position, and matching of the sensor and the installation position is completed according to a one-to-one correspondence between the sensor and the installation position; Complete the configuration and startup at the corresponding installation location according to the installation instructions of the sensor, and obtain the real-time data of the corresponding sensor in real time.
3. The data-driven model building method in an industrial field according to claim 1, characterized in that: The historical data of the target device is obtained according to the historical database, including: Connect to the historical database according to the preset basic database connection information: Select one or more parameter types as the filtering conditions, and obtain the filtering statement conditions based on the parameter type-parameter code mapping table; A filter statement is generated according to the filter statement conditions and imported into a historical database to obtain historical data corresponding to the parameter type.
4. The method for constructing a model based on data-driven in an industrial field according to claim 1, further comprising: The historical data and real-time data of any parameter type are processed into a unified unit based on a preset standard unit to obtain first real-time data and first historical data; Based on the standard time interval, the first real-time data and the first historical data are time aligned to align the starting points, and then the time of all the first real-time data and the first historical data is time standardized according to the standard time interval, and the second data point corresponding to the standardized time node is processed according to the alignment function to obtain standardized real-time data and standardized historical data of the parameter type.
5. The method for constructing a model based on data-driven in an industrial field according to claim 1, characterized in that: According to the industry standards, the resource library is built in combination with the expert database information, including: Obtaining industry specifications and standards for the target device and determining a basic framework of a resource model for the target device; According to the specific professional application information of the target device in the target device matching expert database, the business parameters of the resource model basic framework are adjusted, and the resource model is constructed; Build a resource library and connect to the business data source to obtain original resource data, generate standard resource data after processing through the resource model, and store the resource data of the target device in the resource library.
6. The method for constructing a model based on data-driven in an industrial field according to claim 1, characterized in that: The built-in model is built according to the built-in algorithm of the parameters of the target device, and the third-party AI service is called to build the AI model. The basic model is generated by the built-in model and the AI model, including: Obtain all process parameters of the target device, and select necessary process parameters of the target device based on the functional requirements of the target device: ; Indicates the necessity of any process parameter. represents the standard deviation of the mean of the process parameter and the standard process threshold, represents the standardized historical data of the process parameter at the i-th time node, represents the standard data of the i-th time node of the process parameter, ( )express and dot product, represents the variance of the standardized historical data of the process parameter, n represents that there are n standardized historical data of the process parameter in total, represents the maximum value of the standardized historical data of the process parameter, represents the minimum value of the standardized historical data of the process parameter; Screening the degree of necessity of process parameters according to a minimum degree of necessity screening threshold, and the degree of necessity greater than the minimum degree of necessity is the necessary process parameter of the target device; Selecting a built-in statistical formula according to the necessary process parameters of the target device, verifying the validity of the built-in statistical formula and adjusting the configuration coefficients of the formula parameters of the necessary process parameters to complete the configuration of the built-in model; Design the interface of third-party AI services and allow custom request parameters. Use the request parameters to call the AI model to obtain AI calculation results. The basic model of the target device is jointly constructed by the AI model and the built-in model, supports real-time comparison of AI calculation results and built-in model calculation results, and supports free switching between the two models.
7. The method for constructing a model based on data-driven in the industrial field according to claim 4, characterized in that: Generate a training set based on the historical data of the target device and the resource instance obtained by matching the target device to the resource library, and use the training set to train the basic model to iteratively generate the first model, including: Acquire standardized historical data of the target device and extract features of the historical data of any parameter type; Match resource instances whose feature matching distance with historical data of parameter types is greater than the clustering standard from the resource library, and generate resource instance clusters, and use the resource instance clusters as training set units of corresponding parameter types; The training set units of all parameter types are obtained to generate a training set, and the basic model is iteratively trained based on the training set to generate a first model.
8. The method for constructing a model based on data-driven in the industrial field according to claim 7, characterized in that: The real-time data of the target device is used as a test set, the test set data is substituted into the first model to generate a prediction result, the prediction result is compared with the actual result of the target device, and the model index of the first model is adjusted to obtain a final model, including: Acquire standardized real-time data of the target device, use the standardized real-time data as a test set, and substitute the test set into the first model to generate a prediction result; The prediction results are split according to the parameter type to obtain the prediction parameter results, and the prediction parameter results are compared with the actual results of the corresponding parameter types of the real-time data to obtain the deviation value; If the deviation value is greater than the adjustment threshold, the model indicators of the first model are adjusted according to the deviation value to obtain the final model indicators, and the first model is retrained using the final model indicators to generate the final model. If the deviation value is less than the adjustment threshold, the first model is set as the final model.
9. The method for constructing a model based on data-driven in the industrial field according to claim 8, characterized in that: include: The final model and its parameters are documented and ready to be deployed in a real-world environment for real-time predictions.
10. The method for constructing a model based on data-driven in the industrial field according to claim 8, characterized in that: include: The first model retraining triggers the retraining mechanism by detecting whether the final model indicators have changed.