Method for solving lightweight processing of digital twin model

Abstract

The invention provides a method for solving the lightweight processing of a digital twin model, and the method comprises the steps: screening key data related to a target object from input data, and obtaining the screened data; modeling the screened data to generate an initial digital twinborn model; performing compression processing on the initial digital twinborn model to obtain a compressed digital twinborn model; performing optimization processing on the compressed digital twinborn model to obtain a lightweight digital twinborn model; and executing real-time operation of the target object through the lightweight digital twin model.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular, to a method for solving the lightweight processing of digital twin models. Background Art

[0002] Digital twin technology faces major challenges in processing massive amounts of data. Traditional models often need to process a large amount of redundant information during the construction process, resulting in huge consumption of computing resources and low model efficiency. At the same time, it is difficult to balance the accuracy and real-time performance of the model, especially in rapidly changing environments such as industrial production. How to achieve lightweight while ensuring model accuracy has become a key issue. In addition, the adaptability and scalability of the model are also under test, and it is difficult to adjust and optimize in a timely manner according to the actual operating conditions. These problems are interrelated, forming a complex technical contradiction: the balance between the comprehensiveness and lightweight of the model, the trade-off between accuracy and real-time performance, and the choice between the generality and pertinence of the model. In practical applications, how to extract the most valuable information from massive data, how to retain key features to the greatest extent during model compression, and how to design a dynamic model system that can self-adjust and evolve are all technical problems that need to be solved urgently. The solution to these problems is crucial for enhancing the practical application value of digital twin technology in industrial production and directly affects the accuracy and efficiency of intelligent decision-making and predictive analysis. Summary of the Invention

[0003] The present invention provides a method for solving the lightweight processing of digital twin models, mainly including: Filter out the key data related to the target object from the input data to obtain the filtered data; model the filtered data to generate an initial digital twin model; perform compression processing on the initial digital twin model to obtain a compressed digital twin model; perform optimization processing on the compressed digital twin model to obtain a lightweight digital twin model; perform real-time operations on the target object through the lightweight digital twin model. Further, filtering out the key data related to the target object from the input data includes: obtaining the input data, where the input data includes at least one of historical data, real-time data, and sensor data; for the input data, determining the characteristic parameters related to the target object; extracting key data from the input data according to the characteristic parameters to obtain the filtered data. Further, modeling the filtered data to generate an initial digital twin model includes: obtaining the filtered data; preprocessing the filtered data to obtain formatted data; processing the formatted data through a preset modeling tool to generate an initial digital twin model. Further, performing compression processing on the initial digital twin model to obtain a compressed digital twin model includes: obtaining the initial digital twin model; identifying redundant data from the initial digital twin model; processing the initial digital twin model through a preset compression algorithm to remove the redundant data and optimize the model structure to obtain a compressed digital twin model. Further, performing optimization processing on the compressed digital twin model to obtain a lightweight digital twin model includes: obtaining the compressed digital twin model; performing computational acceleration processing on the compressed digital twin model through a fast algorithm to obtain a first optimized model; processing the first optimized model through parallel computing to obtain a second optimized model; processing the second optimized model through hardware acceleration technology to obtain a lightweight digital twin model. Further, performing real-time operations on the target object through the lightweight digital twin model includes: obtaining the lightweight digital twin model; analyzing the real-time data of the target object through the lightweight digital twin model to obtain an analysis result; performing real-time monitoring operations on the target object according to the analysis result; generating prediction data of the target object according to the analysis result. Further, performing real-time operations on the target object through the lightweight digital twin model includes: obtaining verification data, where the verification data includes actual data or simulation data; verifying the lightweight digital twin model through the verification data to obtain a verification result; adjusting the lightweight digital twin model according to the verification result to obtain an adjusted digital twin model; performing real-time operations on the target object through the adjusted digital twin model.

[0004] The technical solution provided by the embodiment of the present invention may include the following beneficial effects: The present invention discloses a method for constructing and optimizing a lightweight digital twin model based on massive data. This method screens key data from the massive data of the target object, performs preprocessing and modeling to obtain an initial digital twin model. Subsequently, the model is compressed, optimized and lightweight processed to improve the model efficiency. The present invention also introduces a verification data set to verify the model and deploys the verified model to the production environment. In practical applications, the present invention can generate prediction results and decision instructions according to real-time monitoring data, and continuously adjust the model through feedback data to achieve accurate analysis of the operating state of the target object. This method effectively solves the challenges of traditional digital twin models in aspects such as massive data processing, model optimization and real-time application, improves the practicality and efficiency of the model, and provides strong support for intelligent decision-making and predictive analysis in industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] Figure 1 It is a flowchart of a method for solving the lightweight processing of a digital twin model according to the present invention.

[0006] Figure 2 It is a schematic diagram of a method for solving the lightweight processing of a digital twin model according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0007] As Figure 1-2 , a method for solving the lightweight processing of a digital twin model in this embodiment may specifically include: S101. Obtain a massive data set of the target object, where the massive data set includes historical data, real-time data and sensor data, and screen out key data related to the target object from the massive data set to obtain a key data set.

[0008] Obtain a massive data set of the target object to obtain an initial data set including historical data, real-time data and sensor data. According to the source attributes of the initial data set, determine whether the data is related to the target object. If the data is related to the target object, extract key data through a screening mechanism to obtain a preliminary screened data set. Use preprocessing techniques to clean the preliminary screened data set to obtain a data set with consistent formats. Perform structured adjustment on the cleaned data set through formatting processing to determine a standardized key data set. According to the time attributes of the key data set, judge the distribution of historical data and real-time data. If the distribution is uneven, adjust the key data set through a data balancing algorithm to obtain a balanced data set. Obtain the association rules between sensor data and the target object, and judge whether the balanced data set is complete. Optimize the integrity by supplementing missing data to obtain the final key data set.

[0009] Exemplarily, an initial data set is obtained from the massive data collection of the target object, including historical data, real-time data, and sensor data. For example, the operation records of the past year, current state parameters, and sensor data such as temperature and pressure are extracted from industrial equipment. The relevance is judged according to the data source attributes. For example, records with the same target device model are screened out through a label matching algorithm to exclude irrelevant data. If the data is relevant to the target object, a screening mechanism based on a rule engine is used to extract key data. For example, rules are set to filter out outliers and retain data within the normal operating range of the device to obtain a preliminary screened data set. Preprocessing techniques are used to clean the data. For example, missing value imputation algorithms are used to fill in blank fields, and anomaly detection models are used to remove noisy data to ensure consistent data formats. The cleaned data set is structurally adjusted through formatting processing. For example, timestamps are unified into the ISO 8601 standard, and sensor data is converted into the JSON format to determine a standardized key data set. The distribution of historical data and real-time data is judged according to the time attribute. For example, the frequency of data in the past year and the sampling interval of real-time data are counted to analyze whether there is imbalance in the time dimension. If the distribution is uneven, a data balancing algorithm is used to adjust the data set. For example, the SMOTE algorithm is used to oversample the samples in sparse time periods to obtain a balanced data set. The association rules between sensor data and the target object are obtained. For example, the correlation coefficient between temperature and equipment failure rate is calculated through an association analysis algorithm to judge whether the data set is complete. The integrity is optimized by supplementing missing data. For example, missing sensor readings are filled based on a time series prediction model to obtain a final key data set.

[0010] S102. Process the key data set using a preset preprocessing rule to obtain a formatted key data set.

[0011] Obtain a key data set, conduct a preliminary screening on it using a preset preprocessing rule to obtain a preliminarily processed data set. Based on the preliminarily processed data set, use data cleaning techniques to remove redundant items to obtain a cleaned data set. Through the cleaned data set, use standardization rules to adjust the data format to obtain a data set with consistent formats. Based on the data set with consistent formats, use digital twin modeling software to extract key features to obtain a featureized data set. Through the featureized data set, use data mapping techniques to generate an initial digital twin model to obtain a modeled data set. Based on the modeled data set, use a compression algorithm to optimize it to obtain a compressed data set. Through the compressed data set, use lightweight rules to adjust the model structure to obtain a lightweight digital twin model. Based on the lightweight digital twin model, use real-time monitoring techniques to generate a dynamic data stream to obtain a monitored data set. Through the monitored data set, use prediction algorithms to analyze data trends to obtain a prediction result data set.

[0012] Exemplarily, obtain a key data set and perform a preliminary screening on it using a preset preprocessing rule. For example, filter out records with a missing value exceeding 20% from 1000 pieces of raw data collected from production equipment, and obtain a preliminarily processed data set containing 800 records. According to the preliminarily processed data set, use data cleaning technology to remove redundant items. For example, use a rule-based deduplication algorithm to delete duplicate records, and obtain a cleaned data set containing 750 records. Through the cleaned data set, use a standardization rule to adjust the data format. For example, uniformly convert timestamps to the ISO 8601 format and normalize numerical data to between 0 and 1, and obtain a data set with a consistent format. According to the data set with a consistent format, use a digital twin modeling software to extract key features. For example, use the principal component analysis (PCA) method to extract the first 10 principal components, and obtain a characterized data set. Through the characterized data set, use a data mapping technology to generate an initial digital twin model. For example, use a 3D modeling tool to map the feature data into an initial model containing 100 nodes, and obtain a modeled data set. According to the modeled data set, use a compression algorithm to optimize it. For example, use the LZ77 algorithm to compress the model data to 30% of the original size, and obtain a compressed data set. Through the compressed data set, use a lightweight rule to adjust the model structure. For example, reduce the number of polygons in the model by 50%, and obtain a lightweight digital twin model. According to the lightweight digital twin model, use a real-time monitoring technology to generate a dynamic data stream. For example, collect temperature and pressure data every second through sensors and map them into the model, and obtain a monitoring data set. Through the monitoring data set, use a prediction algorithm to analyze the data trend. For example, use the ARIMA model to predict the temperature change in the next 10 minutes, and obtain a prediction result data set.

[0013] S103. Model the formatted key data set through a digital twin modeling tool to obtain an initial digital twin model.

[0014] Process the formatted key data set through digital twin modeling software to obtain the preliminary data mapping relationship. Perform structured modeling based on the preliminary data mapping relationship to determine the framework of the initial digital twin model. Perform parametric assignment on the key data through the framework of the initial digital twin model to obtain the parametric digital twin model. Perform data compression processing based on the parametric digital twin model to obtain the compressed model data. Perform optimization calculation through the compressed model data to obtain the lightweight digital twin model. If the lightweight digital twin model is applicable to the mechanical field, extract its geometric features and judge the adapted mechanical model structure. If the lightweight digital twin model is applicable to the energy field, perform energy flow analysis on it to determine the energy distribution parameters. Perform multi-scenario adaptation tests based on the lightweight digital twin model to obtain the compatibility data of different types of models. Adjust the model through the compatibility data to obtain the final version applicable to multiple digital twin models.

[0015] Exemplarily, process the formatted key data set through digital twin modeling software, use the clustering algorithm to divide the data into five categories to obtain the preliminary data mapping relationship. Perform structured modeling based on the preliminary data mapping relationship, use the hierarchical modeling method to divide the data into the physical layer, functional layer, and behavior layer to determine the framework of the initial digital twin model. Perform parametric assignment on the key data through the framework of the initial digital twin model, use the least squares method to fit the data to obtain the parametric digital twin model. Perform data compression processing based on the parametric digital twin model, use the principal component analysis method to reduce the dimension from 100 to 20 to obtain the compressed model data. Perform optimization calculation through the compressed model data, use the genetic algorithm to optimize the model parameters to obtain the lightweight digital twin model. If the lightweight digital twin model is applicable to the mechanical field, extract its geometric features, use the 3D meshing algorithm to extract the key geometric features and judge the adapted mechanical model structure. If the lightweight digital twin model is applicable to the energy field, perform energy flow analysis on it, use the energy balance equation to calculate the energy distribution of each node to determine the energy distribution parameters. Perform multi-scenario adaptation tests based on the lightweight digital twin model, use the Monte Carlo simulation method to generate 1000 groups of scenario data to obtain the compatibility data of different types of models. Adjust the model through the compatibility data, use the gradient descent method to optimize the model parameters to obtain the final version applicable to multiple digital twin models.

[0016] S104. Remove redundant data from the initial digital twin model and optimize the model structure to obtain the compressed digital twin model.

[0017] By analyzing the data structure of the initial digital twin model, determine the distribution locations of redundant data. According to the determined distribution locations of redundant data, use a data screening algorithm to remove duplicates and obtain the initially refined model data. Through feature extraction of the initially refined model data, obtain the set of key data features. According to the obtained set of key data features, use a clustering analysis method to judge the redundant parts in the model structure. Through trimming the judged redundant parts, obtain the model framework with optimized structure. According to the model framework with optimized structure, use a compression algorithm to encode the data and obtain the compressed digital twin model. Through decoding and verifying the compressed digital twin model, determine the integrity and accuracy of the model. According to the determined integrity and accuracy of the model, use digital twin modeling software to adjust the parameters of the compressed model and obtain the adjusted lightweight model. By matching the adjusted lightweight model with the actual production data, obtain the model parameters required for real-time monitoring.

[0018] Exemplarily, by analyzing the data structure of the initial digital twin model, use the principal component analysis (PCA) method to determine the distribution locations of redundant data. For example, identify 80% of the duplicate data in the model. According to the determined distribution locations of redundant data, use a hash algorithm to screen the data, remove duplicates, and obtain the initially refined model data, with the data volume reduced by 60%. Through feature extraction of the initially refined model data, use a convolutional neural network (CNN) to extract the set of key data features. For example, extract the core features accounting for 20% in the model. According to the obtained set of key data features, use the K-means clustering analysis method, set the number of clusters to 5, and judge the redundant parts in the model structure. For example, identify 30% of the redundant nodes. Through trimming the judged redundant parts, use a pruning algorithm to remove the redundant nodes and obtain the model framework with optimized structure, with the model complexity reduced by 40%. According to the model framework with optimized structure, use the Huffman coding algorithm to compress and encode the data, with a compression rate of 50%, and obtain the compressed digital twin model. Through decoding and verifying the compressed digital twin model, use the mean squared error (MSE) to evaluate the integrity and accuracy of the model, with the error controlled within 0.01. According to the determined integrity and accuracy of the model, use digital twin modeling software to adjust the parameters of the compressed model. For example, adjust the learning rate to 0.001 and obtain the adjusted lightweight model. By matching the adjusted lightweight model with the actual production data, use real-time data stream processing technology to obtain the model parameters required for real-time monitoring. For example, process 1000 data items per second.

[0019] S105. Optimize the compressed digital twin model by using a fast algorithm and parallel computing technology to obtain a lightweight digital twin model.

[0020] Obtain the compressed digital twin model and extract the structural and parameter information of the model. Simplify the structure of the model using a fast algorithm to remove redundant nodes and connections. Optimize the parameters of the simplified model and use the gradient descent method to adjust the parameter values. Introduce parallel computing technology, decompose the computing tasks of the model into multiple subtasks. Allocate the subtasks to multiple computing units and accelerate the model processing through parallel computing. Integrate the results of parallel computing to generate an optimized intermediate model. According to the actual production requirements, evaluate the accuracy of the intermediate model to determine whether it meets the real-time requirements. If the accuracy evaluation fails, iterate and optimize the model, adjusting the algorithm parameters and computing unit allocation. Obtain a lightweight digital twin model that meets the real-time requirements and complete the optimization process.

[0021] Exemplarily, obtain the compressed digital twin model and extract the structural and parameter information of the model, specifically including the number of nodes, connection relationships, and weight values. Simplify the structure of the model using a fast algorithm, remove redundant nodes and connections through a pruning algorithm, for example, reduce the number of nodes from 1000 to 500. Optimize the parameters of the simplified model, use the gradient descent method to adjust the parameter values, set the learning rate to 0.01, and iterate 100 times to minimize the loss function. Introduce parallel computing technology, decompose the computing tasks of the model into multiple subtasks, for example, decompose matrix operations into 4 sub-matrices. Allocate the subtasks to multiple computing units and accelerate the model processing through GPU parallel computing, shortening the computing time from 10 seconds to 2 seconds. Integrate the results of parallel computing to generate an optimized intermediate model, ensuring the consistency and integrity of the computing results. According to the actual production requirements, evaluate the accuracy of the intermediate model, use the mean square error metric to determine whether it meets the real-time requirements, and set the error threshold to 0.001. If the accuracy evaluation fails, iterate and optimize the model, adjusting the algorithm parameters and computing unit allocation, for example, adjust the learning rate to 0.005 and increase the number of iterations to 150 times. Obtain a lightweight digital twin model that meets the real-time requirements and complete the optimization process. The storage space of the final model is reduced by 50%, and the computing efficiency is increased by 80.

[0022] S106. Obtain a validation data set from the actual production environment. The validation data set includes actual data and simulated data. Validate the lightweight digital twin model through the validation data set to obtain a validated digital twin model.

[0023] Collect multi-source data from the actual production environment, extract key data related to the target object through data screening methods, and obtain a preliminary data set. According to the preliminary data set, use a simulation algorithm to generate simulated data corresponding to the actual data, and obtain a verification data set containing actual data and simulated data. Model the key data in the verification data set through digital twin modeling software to obtain an initial digital twin model. According to the initial digital twin model, use a lightweight processing method to optimize the data storage and calculation requirements to obtain a lightweight digital twin model. Verify the lightweight digital twin model for the first time through the actual data in the verification data set to obtain a preliminary verification result. According to the preliminary verification result, if the deviation between the model output and the actual data exceeds the threshold, optimize it by adjusting the model parameters to obtain an adjusted lightweight digital twin model. Verify the adjusted lightweight digital twin model for the second time through the simulated data in the verification data set to obtain a simulation verification result. According to the simulation verification result, if the consistency between the model output and the simulated data is lower than the standard, determine the error source through data comparison and analysis to obtain an error correction plan. Finally, adjust the lightweight digital twin model according to the error correction plan to obtain a verified digital twin model.

[0024] Exemplarily, multi-source data is collected from the actual production environment, such as sensor data, device operation logs, and environmental parameters. Key data related to the target object is extracted through a data screening method. For example, vibration data with a frequency higher than 10 Hz is screened out to obtain a preliminary data set. Based on the preliminary data set, simulation algorithms are used to generate simulation data corresponding to the actual data. For example, the Gaussian distribution algorithm is used to generate temperature simulation data with a mean of 50 and a standard deviation of 5, resulting in a validation data set that includes actual data and simulation data. The key data in the validation data set is modeled through digital twin modeling software. For example, ANSYS software is used to perform three-dimensional modeling of the device operation state to obtain an initial digital twin model. Based on the initial digital twin model, a lightweight processing method is adopted to optimize the data storage and calculation requirements. For example, the PCA algorithm is used to reduce the data dimension from 100 dimensions to 10 dimensions, resulting in a lightweight digital twin model. The lightweight digital twin model is first verified through the actual data in the validation data set. For example, the temperature data output by the model is compared with the actual temperature data to obtain a preliminary verification result. According to the preliminary verification result, if the deviation between the model output and the actual data exceeds the threshold, such as the temperature deviation exceeds ±2°C, the model parameters are adjusted for optimization. For example, the heat conduction coefficient is adjusted from 0.5 to 0.6, resulting in an adjusted lightweight digital twin model. The adjusted lightweight digital twin model is secondarily verified through the simulation data in the validation data set. For example, the pressure data output by the model is compared with the simulated pressure data to obtain a simulation verification result. According to the simulation verification result, if the consistency between the model output and the simulation data is lower than the standard, such as the correlation coefficient is lower than 0.9, the error source is determined through data comparison and analysis. For example, it is found that the error mainly comes from unreasonable initial parameter settings, resulting in an error correction plan. The lightweight digital twin model is finally adjusted through the error correction plan. For example, the initial parameters are reset and remodeled to obtain a verified digital twin model.

[0025] S107. Deploy the verified digital twin model to the production environment, obtain real-time monitoring data, and determine a prediction result and a decision instruction based on the real-time monitoring data.

[0026] Obtain real-time data in the production environment through the lightweight digital twin model to get the real-time monitoring data stream. Perform data preprocessing on the real-time monitoring data stream to determine the standardized monitoring data set. Input the standardized monitoring data set into the verified digital twin model to obtain the preliminary model calculation results. Compare the model calculation results with the historical data to judge the deviation range of the prediction results. If the deviation range exceeds the preset threshold, dynamically adjust the model parameters to obtain the optimized prediction results. Generate the corresponding decision instruction set according to the optimized prediction results and determine the priority ranking of the instructions. Connect the instruction set after priority ranking with the production environment system to obtain the feedback data after execution. Update the real-time state of the digital twin model according to the feedback data to obtain the dynamic performance indicators of the model. Adjust the data collection frequency through the dynamic performance indicators to determine the monitoring data stream for the next cycle.

[0027] Exemplarily, obtain real-time data in the production environment through the lightweight digital twin model, collect parameters such as temperature and pressure using sensors, with a sampling frequency of once per second, to get the real-time monitoring data stream. Perform data preprocessing on the real-time monitoring data stream, use the Z-score standardization method to normalize the data, map the temperature value to the range of 0 to 1, and determine the standardized monitoring data set. Input the standardized monitoring data set into the verified digital twin model, use the long short-term memory network (LSTM) algorithm for calculation to obtain the preliminary model calculation results. Compare the model calculation results with the historical data, calculate the root mean square error (RMSE) value to be 0.05, and judge the deviation range of the prediction results. If the deviation range exceeds the preset threshold of 0.1, dynamically adjust the model parameters, use the gradient descent method to optimize the learning rate to 0.001, and obtain the optimized prediction results. Generate the corresponding decision instruction set according to the optimized prediction results, use the weighted scoring method to sort the instructions, and determine the priority ranking of the instructions. Connect the instruction set after priority ranking with the production environment system, execute the temperature adjustment instruction, and obtain the feedback data after execution. Update the real-time state of the digital twin model according to the feedback data, calculate the model accuracy to reach 98%, and obtain the dynamic performance indicators of the model. Adjust the data collection frequency through the dynamic performance indicators, reduce the sampling frequency to once every two seconds, and determine the monitoring data stream for the next cycle.

[0028] S108. Adjust the verified digital twin model according to the feedback data in the production environment to obtain the adjusted digital twin model.

[0029] Obtain feedback data in the production environment and determine the time series characteristics of the feedback data. Analyze the output of the lightweight digital twin model through the time series characteristics to judge the deviation between the model prediction value and the feedback data. Calculate the direction of model parameter adjustment according to the deviation to obtain the initial range of parameter adjustment. Use the initial range to perform parameter iteration on the verified digital twin model to obtain the iterated model parameter set. Update the structure of the lightweight digital twin model through the parameter set to determine the preliminary version of the adjusted model. Obtain the real-time output of the preliminary version of the adjusted model in the production environment and judge the consistency between the output and the feedback data. If the consistency is lower than the set threshold, adjust the weight distribution of the model through the consistency difference to obtain the model with optimized weights. Recalculate the prediction results in the production environment according to the model with optimized weights to determine the adjusted digital twin model. Generate real-time monitoring data through the adjusted digital twin model to obtain the feedback data for the next cycle.

[0030] Exemplarily, obtain feedback data in the production environment and extract the time series characteristics of the feedback data. For example, use the sliding window technique to perform segmented analysis on the production data in the past 24 hours. Analyze the output of the lightweight digital twin model through the time series characteristics, and use the mean square error algorithm to calculate the deviation between the model prediction value and the feedback data. For example, the deviation value is 0.15. Calculate the direction of model parameter adjustment according to the deviation, and use the gradient descent algorithm to determine the initial range of parameter adjustment. For example, the learning rate is set to 0.01. Use the initial range to perform parameter iteration on the verified digital twin model, and obtain the iterated model parameter set through 500 iterations. Update the structure of the lightweight digital twin model through the parameter set. For example, adjust the number of neural network layers to 3 layers to determine the preliminary version of the adjusted model. Obtain the real-time output of the preliminary version of the adjusted model in the production environment, and use the Pearson correlation coefficient to judge the consistency between the output and the feedback data. For example, the correlation coefficient is 0.85. If the consistency is lower than the set threshold of 0.9, adjust the weight distribution of the model through the consistency difference, and use the weighted least squares method to obtain the model with optimized weights. Recalculate the prediction results in the production environment according to the model with optimized weights. For example, the prediction accuracy is improved to 92% to determine the adjusted digital twin model. Generate real-time monitoring data through the adjusted digital twin model. For example, collect data once per second to obtain the feedback data for the next cycle.

[0031] S109. Analyze the real-time monitoring data through the adjusted digital twin model to obtain the operating state parameters of the target object.

[0032] Receive real-time monitoring data through the adjusted digital twin model to obtain the original data stream. Preprocess the original data stream to obtain a standardized data format. Input the standardized data format into the lightweight digital twin model to obtain the preliminary feature extraction results. Compress and optimize the preliminary feature extraction results to determine the refined feature parameters. Perform model verification based on the refined feature parameters to determine whether the model output meets the preset accuracy threshold. If the model output meets the preset accuracy threshold, analyze the feature parameters through the verified model to obtain the operating state parameters of the target object. Compare the real-time data based on the operating state parameters to obtain the change trend of the state parameters. Perform correlation analysis on the change trend and historical data to determine the abnormal points of the operating state. Match the abnormal points with the preset rules to obtain the operating state evaluation result of the target object.

[0033] Exemplarily, receive real-time monitoring data through the adjusted digital twin model, such as parameters like temperature, pressure, and vibration collected from sensors, to obtain the original data stream containing timestamps and values. Preprocess the original data stream, use the sliding window algorithm to smooth the data, and remove outliers to obtain a standardized data format, such as uniformly converting the temperature value to degrees Celsius and the pressure value to pascals. Input the standardized data format into the lightweight digital twin model, use a convolutional neural network to extract the spatial features of the data, and obtain the preliminary feature extraction results, such as the gradient feature of temperature change and the spectral feature of pressure fluctuation. Compress and optimize the preliminary feature extraction results, use the principal component analysis method to reduce the high-dimensional features to three dimensions, and determine the refined feature parameters, such as the principal component coefficients of the temperature gradient and the main frequency components of the pressure spectrum. Perform model verification based on the refined feature parameters, use the mean squared error to calculate the deviation between the model output and the real data, and determine whether the model output meets the preset accuracy threshold, such as passing the verification if the error is less than 0.1. If the model output meets the preset accuracy threshold, analyze the feature parameters through the verified model, use the support vector machine algorithm to classify the temperature gradient and pressure spectrum, and obtain the operating state parameters of the target object, such as whether the device is in an overheated or overpressured state. Compare the real-time data based on the operating state parameters, use the time series analysis method to calculate the change trend of the parameters, such as the hourly rising rate of temperature and the minute-by-minute fluctuation amplitude of pressure. Perform correlation analysis on the change trend and historical data, use the clustering algorithm to match the current trend with the historical normal state, and determine the abnormal points of the operating state, such as the temperature rising rate exceeding 20% of the historical normal range. Match the abnormal points with the preset rules, use the decision tree algorithm to classify the abnormal points, and obtain the operating state evaluation result of the target object, such as the device being in a minor or severe fault state.

[0034] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. The present invention has only been described in detail with reference to the preferred embodiments. Those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and all of them should be covered within the scope of the claims of the present invention.

Claims

1. A method for solving the lightweight processing of digital twin models, characterized in that: include: Filter out key data related to the target object from the input data to obtain filtered data; The filtered data is modeled to generate an initial digital twin model; the initial digital twin model is compressed to obtain a compressed digital twin model; the compressed digital twin model is optimized to obtain a lightweight digital twin model; and real-time operations of the target object are performed through the lightweight digital twin model.

2. The method according to claim 1, characterized in that The method of filtering out key data related to the target object from input data includes: obtaining input data, wherein the input data includes at least one of historical data, real-time data and sensor data; determining characteristic parameters related to the target object for the input data; and extracting key data from the input data according to the characteristic parameters to obtain filtered data.

3. The method according to claim 1, characterized in that The modeling of the filtered data to generate an initial digital twin model includes: acquiring the filtered data; preprocessing the filtered data to obtain formatted data; and processing the formatted data by a preset modeling tool to generate an initial digital twin model.

4. The method according to claim 1, characterized in that The compressing the initial digital twin model to obtain a compressed digital twin model includes: acquiring the initial digital twin model; identifying redundant data from the initial digital twin model; processing the initial digital twin model through a preset compression algorithm, removing the redundant data and optimizing the model structure, and obtaining a compressed digital twin model.

5. The method according to claim 1, characterized in that The method of optimizing the compressed digital twin model to obtain a lightweight digital twin model includes: obtaining the compressed digital twin model; performing computational acceleration processing on the compressed digital twin model by a fast algorithm to obtain a first optimized model; processing the first optimized model by parallel computing to obtain a second optimized model; and processing the second optimized model by hardware acceleration technology to obtain a lightweight digital twin model.

6. The method according to claim 1, characterized in that The real-time operation of the target object is performed through the lightweight digital twin model, including: obtaining the lightweight digital twin model; analyzing the real-time data of the target object through the lightweight digital twin model to obtain analysis results; performing real-time monitoring operations on the target object according to the analysis results; and generating prediction data of the target object according to the analysis results.

7. The method according to claim 1, characterized in that The real-time operation of the target object is performed through the lightweight digital twin model, including: obtaining verification data, wherein the verification data includes actual data or simulation data; verifying the lightweight digital twin model through the verification data to obtain a verification result; adjusting the lightweight digital twin model according to the verification result to obtain an adjusted digital twin model; and performing real-time operation of the target object through the adjusted digital twin model.

Images