Rectified product concentration on-line detection system

Abstract

The invention belongs to the field of rectification, and particularly relates to a rectification product concentration online detection system which comprises a data acquisition module, the data acquisition module is connected with a data processing and fusion module, the data processing and fusion module is connected with a digital twinborn model module, the digital twinborn model module is connected with a model self-adaptive correction module, and the model self-adaptive correction module is connected with a digital twinborn model. The model adaptive correction module is connected with a working condition mode identification module, the working condition mode identification module is connected with a model migration loading module, the model migration loading module is connected with a multi-working condition model library management module, and the multi-working condition model library management module is connected with a sensor data module. According to the system, high-real-time and high-precision concentration detection is achieved, the strong self-adaption and working condition migration capacity is achieved, the robustness and reliability of the system are remarkably improved, and intelligent self-management and self-recovery of the whole life cycle are achieved.

Description

Technical Field

[0001] This invention relates to the field of distillation technology, and in particular to an online detection system for the concentration of distillation products. Background Technology

[0002] Distillation, as a critical and energy-intensive separation unit operation in process industries such as petroleum, chemical, and pharmaceutical, relies heavily on real-time and accurate detection of product concentration for process optimization, energy conservation, and high-quality control. Currently, online detection and control of distillation product concentration in industry primarily relies on the following technologies:

[0003] The first category comprises direct measurement technologies based on physical analysis instruments. These mainly include online industrial chromatographs and online near-infrared spectrometers. These instruments can directly measure the concentration of product components and were once considered the "gold standard" for concentration detection. However, they have inherent and insurmountable drawbacks:

[0004] Significant measurement lag: These instruments typically require sampling, preprocessing, and analysis cycles, resulting in measurement results lagging behind the actual process by several minutes or even longer. This large pure lag time severely compromises the real-time performance of the control system, making it difficult to effectively apply advanced control systems and even causing fluctuations in product quality.

[0005] High maintenance costs and reliability issues: Instruments come into direct contact with process media and are susceptible to material contamination, corrosion, crystallization, etc., requiring frequent calibration, maintenance, and component replacement. Their purchase and life-cycle maintenance costs are high. In the event of a malfunction, a "blind spot" in production monitoring will be created, posing a significant risk to production safety and product quality.

[0006] The second category is indirect inference techniques based on soft sensing. Soft sensing techniques emerged to overcome the lag problem of physical instruments. They infer concentration values ​​in real time by establishing mathematical models between easily measurable process variables (such as temperature and pressure) and difficult-to-measure product concentrations. However, traditional soft sensing techniques face significant challenges in practice:

[0007] Poor model adaptability: Whether based on mechanisms or data, the parameters of the model are usually fixed under specific operating conditions. When the properties of raw materials change, product specifications are switched, tray efficiency slowly decreases due to fouling, or production load is significantly adjusted, the model's prediction accuracy will deteriorate significantly. It lacks the "self-learning" ability to track dynamic changes in the process, requiring engineers to frequently recalibrate manually, greatly reducing its practicality.

[0008] High dependence on and vulnerability to critical sensors: The input of the soft measurement model is heavily dependent on a few critical process sensors (such as the temperature of the sensitive plate). Once these sensors drift or fail, it will lead to a "garbage in, garbage out" situation, and the output of the entire soft measurement system will become completely unreliable. Moreover, the system often lacks the ability to identify such front-end data failures.

[0009] The third category is digital twin applications under single operating conditions. In recent years, with the development of modeling technology, digital twin models based on rigorous mechanisms have been attempted for process monitoring. However, existing applications are mostly limited to single, stable, ideal operating conditions. When faced with frequent operating condition changes in actual production, the model requires a long re-initialization and convergence time, which cannot meet the rapid response requirements of continuous production. In addition, these models usually operate as a "black box," lacking quantification of the uncertainty of their output results, and also lacking version management and self-recovery mechanisms when the model's performance degrades, resulting in insufficient robustness for industrial applications.

[0010] In summary, existing technical solutions have significant shortcomings in terms of real-time performance, reliability, adaptability, and engineering practicality. Summary of the Invention

[0011] The purpose of this invention is to address the shortcomings of existing technologies in terms of real-time performance, reliability, adaptability, and engineering practicality by proposing an online detection system for distillation product concentration.

[0012] To achieve the above objectives, the present invention adopts the following technical solution:

[0013] An online concentration detection system for distillation products includes a data acquisition module connected to a data processing and fusion module, a digital twin model module connected to the data acquisition module, a model adaptive calibration module connected to the model adaptive calibration module, a working condition pattern recognition module connected to the working condition pattern recognition module, a model migration and loading module connected to the model migration and loading module, a multi-working condition model library management module connected to the multi-working condition model library management module, a sensor data module connected to a key sensor fault diagnosis module, a fault data reconstruction module connected to the key sensor fault diagnosis module, a system health module connected to the fault data reconstruction module, the digital twin model module further connected to a prediction uncertainty quantification module, a prediction uncertainty quantification module connected to a digital twin management module, a closed-loop control interface module connected to a closed-loop control interface module, an edge-cloud collaborative computing module connected to the edge-cloud collaborative computing module, and a dynamic performance evaluation module.

[0014] Preferably, the data acquisition module is used to acquire real-time dynamic process data of the distillation process and discontinuous direct concentration data from at least one physical analysis instrument;

[0015] The data processing and fusion module is used to preprocess and time-align the real-time dynamic process data and the discontinuous direct concentration data to form a multimodal dataset.

[0016] Preferably, the digital twin model module is constructed based on the distillation process mechanism, receives real-time dynamic process data from the multimodal dataset as input, and outputs real-time product concentration estimates.

[0017] The model adaptive correction module compares the discontinuous direct concentration data with the predicted values ​​of the digital twin model at the corresponding historical time, and dynamically adjusts the internal parameters of the digital twin model based on the comparison results.

[0018] Preferably, the operating condition mode recognition module is used to automatically identify and mark the current operating condition of the distillation column based on the feed components, product specifications and operating parameters, and package and store the stable operating data under the operating condition and the corresponding calibrated digital twin model parameters as a specific operating condition model package.

[0019] The mode migration loading module automatically matches the most similar working condition model package from the historical working condition model library when the system detects or receives a working condition switching command, and loads the model parameters in it as initial values ​​into the digital twin model module to achieve rapid startup and convergence under the new working condition.

[0020] Preferably, the multi-condition model library management module is used to store, index, and manage multiple specific condition model packages, and record the usage frequency and prediction accuracy of each model package to provide priority reference for model matching;

[0021] The sensor data module learns the causal relationships between multiple process variables in the distillation column based on historical data, and constructs a causal graph between the variables to identify abnormal data sources.

[0022] Preferably, the key sensor fault diagnosis module receives the output of the sensor data causal inference network module. When a key sensor data undergoes a sudden change or continuous deviation, it analyzes the logical consistency between the data and other related variables through a causal graph to determine whether the sensor has malfunctioned and triggers an alarm.

[0023] Preferably, when the critical sensor fault diagnosis module confirms a sensor fault, the fault data reconstruction module uses the sensor data causal inference network module to estimate and reconstruct the alternative data value of the faulty sensor in real time based on the data of other normally operating sensors.

[0024] Preferably, the system health module is used to assess the health status of each component of the system; when a sensor fault is diagnosed and data reconstruction is enabled, the system is automatically marked as "degraded operation" mode, and the uncertainty range of the concentration estimate at this time is assessed and reported.

[0025] Preferably, the prediction uncertainty quantification module is connected to the digital twin model module, and outputs not only the point estimate of the concentration, but also the confidence interval of the estimate, providing a risk basis for advanced control;

[0026] The digital twin management module records the parameter versions after each adaptive correction of the model. When the performance of the corrected model deteriorates after evaluation, it supports one-click rollback to the previous stable version to ensure system reliability.

[0027] Preferably, the closed-loop control interface module sends the final product concentration estimate and uncertainty information to the advanced process control or real-time optimization system of the distillation column in real time using a standard communication protocol, in order to form a closed-loop control loop for product quality.

[0028] The edge-cloud collaborative computing module deploys data preprocessing and real-time digital twin computing on the edge to ensure real-time performance, while deploying computationally intensive tasks such as model training and operational model library management on the cloud platform to achieve optimized allocation of computing resources.

[0029] The dynamic performance evaluation module automatically generates system performance reports periodically, compares the long-term deviations between soft measurement values ​​and physical sensor values, evaluates the model's accuracy and stability, and provides maintenance suggestions.

[0030] The beneficial effects of the online concentration detection system for distillation products described in this invention are as follows:

[0031] Achieving high real-time performance and high precision in concentration detection: By constructing a digital twin model module and using it as the core for real-time simulation, this invention can output product concentration estimates in seconds or milliseconds, completely eliminating the measurement lag of several minutes caused by sampling and analysis cycles in traditional physical analysis instruments (such as industrial chromatographs). Simultaneously, the model adaptive calibration module continuously fine-tunes the model using the lagging high-precision physical instrument data, ensuring the accuracy of the soft measurement results in long-term operation and achieving a balance between lag and precision.

[0032] Possessing powerful adaptive and operational condition migration capabilities: This invention solves the problem of sharp accuracy drops in traditional soft measurement models when operational conditions change through the synergistic effect of operational condition pattern recognition, model migration loading, and multi-operational condition model library management. The system can automatically identify operational condition transitions and quickly retrieve matching historical model parameters from the knowledge base for initialization, enabling the digital twin model to converge rapidly under new operational conditions. This greatly improves the system's practicality and stability in variable operational condition production environments and reduces manual intervention.

[0033] Significantly improving system robustness and reliability: By introducing a sensor fault diagnosis and data reconstruction mechanism based on causal inference, this invention effectively addresses the common failure point of soft measurement systems: critical sensor failure. The system can not only diagnose sensor faults and issue alarms in a timely manner, but also reconstruct reliable alternative data through causal networks. This ensures that the digital twin model can continue to provide valuable concentration estimates during sensor failures, avoiding system paralysis caused by "single point failures" and greatly enhancing the system's fault tolerance and operational continuity.

[0034] Achieving intelligent self-management and self-recovery throughout the entire lifecycle: This invention integrates modules such as model version management, system health self-assessment, and dynamic performance assessment, enabling the system to possess self-monitoring, self-assessment, and self-optimization capabilities. The digital twin management module prevents model performance degradation due to improper calibration and supports one-click rollback; the system health module transparently reports operating status and uncertainties, providing support for operator decision-making; and the dynamic performance assessment module provides data support for preventative maintenance, realizing a shift from passive maintenance to proactive management.

[0035] Optimizing computing resources and supporting advanced control applications: Through the edge-cloud collaborative computing module, this invention rationally allocates the computing load, placing tasks with high real-time requirements at the edge and computationally intensive tasks in the cloud, ensuring the system's timely response and cost-effectiveness. The closed-loop control interface module transmits high-real-time concentration estimates and their confidence intervals to the advanced control system, laying a solid data foundation for achieving accurate and reliable product quality closed-loop control and real-time optimization (RTO), ultimately achieving the goals of energy saving, consumption reduction, quality improvement, and efficiency enhancement.

[0036] Providing quantified uncertainty to support risk-based decision-making: The output of the uncertainty quantification module for prediction changes the limitation of traditional soft sensing, which only provides a "definite value." It provides the control system with a confidence interval for the concentration estimate, enabling advanced control algorithms to assess the risk of control decisions. While pursuing optimization objectives, it proactively avoids control action risks caused by model prediction uncertainties, thereby improving the intelligence level of decision-making and operational safety of the entire production process. Attached Figure Description

[0037] Figure 1 This is a block diagram of an online concentration detection system for distillation products proposed in this invention;

[0038] Figure 2 This is a flowchart of the data processing and fusion module of an online distillation product concentration detection system proposed in this invention;

[0039] Figure 3 This is a flowchart of a digital twin model module for an online distillation product concentration detection system proposed in this invention;

[0040] Figure 4 This is a flowchart of the operating mode recognition module of an online distillation product concentration detection system proposed in this invention;

[0041] Figure 5 This is a flowchart of the multi-condition model library management module of an online distillation product concentration detection system proposed in this invention. Detailed Implementation

[0042] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0043] Example 1

[0044] Reference Figures 1-5 An online concentration detection system for distillation products includes a data acquisition module, which is connected to a data processing and fusion module. The data processing and fusion module is connected to a digital twin model module, which is connected to a model adaptive calibration module. The model adaptive calibration module is connected to a working condition pattern recognition module, which is connected to a model migration and loading module. The model migration and loading module is connected to a multi-working condition model library management module, which is connected to a sensor data module. The sensor data module is connected to a key sensor fault diagnosis module, which is connected to a fault data reconstruction module. The fault data reconstruction module is connected to a system health module. The digital twin model module is also connected to a prediction uncertainty quantification module, which is connected to a digital twin management module. The digital twin management module is connected to a closed-loop control interface module, which is connected to an edge-cloud collaborative computing module. The edge-cloud collaborative computing module is connected to a dynamic performance evaluation module.

[0045] In this embodiment, the data acquisition module is used to acquire real-time dynamic process data of the distillation process and discontinuous direct concentration data from at least one physical analysis instrument.

[0046] The data processing and fusion module is used to preprocess and time-align real-time dynamic process data and discontinuous direct concentration data to form a multimodal dataset.

[0047] In this embodiment, the digital twin model module is built based on the distillation process mechanism, receives real-time dynamic process data from the multimodal dataset as input, and outputs real-time product concentration estimates.

[0048] The model adaptive calibration module compares discontinuous direct concentration data with the predicted values ​​of the digital twin model at the corresponding historical time, and dynamically adjusts the internal parameters of the digital twin model based on the comparison results.

[0049] In this embodiment, the operating condition mode recognition module is used to automatically identify and mark the current operating condition of the distillation column based on the feed components, product specifications and operating parameters, and package and store the stable operating data under the operating condition and the corresponding calibrated digital twin model parameters as a specific operating condition model package.

[0050] The mode migration loading module automatically matches the most similar working condition model package from the historical working condition model library when the system detects or receives a working condition switching command, and loads the model parameters in it as initial values ​​into the digital twin model module to achieve rapid startup and convergence under the new working condition.

[0051] In this embodiment, the multi-condition model library management module is used to store, index, and manage multiple specific condition model packages, and record the usage frequency and prediction accuracy of each model package to provide priority reference for model matching;

[0052] The sensor data module learns the causal relationships between multiple process variables in the distillation column based on historical data, and constructs a causal graph between the variables to identify abnormal data sources.

[0053] In this embodiment, the key sensor fault diagnosis module receives the output of the sensor data causal inference network module. When a key sensor data changes abruptly or deviates continuously, it analyzes the logical consistency between the data and other related variables through a causal graph to determine whether the sensor has malfunctioned and triggers an alarm.

[0054] In this embodiment, when the critical sensor fault diagnosis module confirms that a certain sensor is faulty, the fault data reconstruction module uses the sensor data causal inference network module to estimate and reconstruct the alternative data value of the faulty sensor in real time based on the data of other normally operating sensors.

[0055] In this embodiment, the system health module is used to assess the health status of each component of the system; when a sensor fault is diagnosed and data reconstruction is enabled, the system is automatically marked as "degraded operation" mode, and the uncertainty range of the concentration estimate at this time is assessed and reported.

[0056] In this embodiment, the prediction uncertainty quantification module is connected to the digital twin model module. It not only outputs the point estimate of the concentration, but also the confidence interval of the estimate, providing a risk basis for advanced control.

[0057] The digital twin management module records the parameter versions after each adaptive correction of the model. When the performance of the corrected model deteriorates after evaluation, it supports one-click rollback to the previous stable version to ensure system reliability.

[0058] In this embodiment, the closed-loop control interface module sends the final product concentration estimate and uncertainty information to the advanced process control or real-time optimization system of the distillation column in real time using a standard communication protocol, in order to form a closed-loop control loop for product quality.

[0059] The edge-cloud collaborative computing module deploys data preprocessing and real-time digital twin computing on the edge to ensure real-time performance, while deploying computationally intensive tasks such as model training and operational model library management on the cloud platform to achieve optimized allocation of computing resources.

[0060] The dynamic performance evaluation module automatically generates system performance reports periodically, compares the long-term deviations between soft measurement values ​​and physical sensor values, evaluates the model's accuracy and stability, and provides maintenance suggestions.

[0061] Example 2

[0062] The difference between this embodiment and Embodiment 1 is that it also includes an anti-external interference module. This anti-external interference module is connected between the data processing and fusion module and the digital twin model module. It is used to: monitor and identify uncontrollable external interferences such as sudden changes in feed temperature / pressure and sudden drops in ambient temperature in real time; establish an interference transmission model to predict the impact path and magnitude of the interference on the steady state of the distillation column; perform feedforward compensation on the process data input to the digital twin model in the early stage of interference; and work with the model adaptive correction module to distinguish between parameter drift and external interference to avoid miscorrection.

[0063] The rest is the same as in Example 1.

[0064] Example 3

[0065] The difference between this embodiment and Embodiment 1 is that it also includes a model incremental learning and knowledge distillation module. This module is connected between the model adaptive correction module and the multi-condition model library management module. It is used to: incrementally learn from continuously generated correction data to avoid the computational burden of model retraining; use knowledge distillation technology to compress the core knowledge of complex digital twin models into lightweight inference models; prioritize loading lightweight models to achieve second-level response when rapid condition switching is required; and establish a model performance degradation early warning mechanism to trigger the full model retraining threshold.

[0066] The rest is the same as in Example 1.

[0067] Test case

[0068] 1. Experimental Objective

[0069] To verify the superiority of the system of this invention over traditional methods, the focus is on evaluating its performance in terms of measurement real-time performance, operating condition adaptability, fault tolerance and long-term stability.

[0070] 2. Test Setup

[0071] Test subject: Benzene-toluene separation distillation column industrial unit

[0072] Comparison System:

[0073] Comparison System A: Traditional Online Industrial Chromatograph

[0074] Comparison System B: Soft Measurement System Based on Temperature-Pressure Regression

[0075] The system of this invention includes all the modules of Embodiments 1, 2, and 3.

[0076] Test period: 30 consecutive days

[0077] Evaluation metrics: Mean Absolute Error (MAE), Response Lag Time, Condition Switching Convergence Time, Fault Recovery Time, System Availability;

[0078] The test data are as follows:

[0079] Table 1: Steady-state accuracy and real-time performance test results

[0080]

[0081] Table 2: Results of Adaptability Test for Operating Condition Switching

[0082]

[0083]

[0084] Table 3: Results of Anti-interference and Fault Recovery Tests

[0085]

[0086]

[0087] Table 4: Statistics on Long-Term Operational Stability

[0088]

[0089] Experimental results analysis: (1) Balance between real-time performance and accuracy

[0090] The system of this invention successfully resolves the contradiction between real-time performance and accuracy in traditional methods. As shown in Table 1, while maintaining measurement accuracy similar to that of industrial chromatographs (MAE 0.18% vs 0.15%), it achieves second-level response and completely eliminates the 4.5-minute measurement lag.

[0091] (2) The ability to adapt to different operating conditions has been significantly improved.

[0092] As can be seen from the data in Table 2, the system of this invention exhibits significant advantages during operating condition switching. Traditional soft measurement systems require manual recalibration and converge slowly, while the system of this invention achieves automatic and rapid convergence through operating condition identification and model transfer. In particular, the incremental learning and knowledge distillation techniques introduced in Example 3 further reduce the convergence time from 85 minutes to 32 minutes.

[0093] (3) Excellent anti-interference and fault tolerance performance

[0094] Table 3 shows that the system of the present invention exhibits extremely strong robustness in the face of external interference and sensor failure. The anti-external interference module (Example 2) effectively suppresses the impact of sudden changes in feed temperature, while the fault diagnosis and data reconstruction mechanism ensures the continuous operation of the system in the event of critical sensor failure.

[0095] (4) Excellent long-term operational stability

[0096] Based on 30 days of long-term operational data (Table 4), the system of this invention is superior to traditional methods in terms of availability, maintenance frequency, and stability. The system availability reaches 99.5%, far exceeding the 87.3% of industrial chromatographs, and no manual maintenance intervention is required throughout the entire testing period.

[0097] Conclusion: Through systematic testing, the online concentration detection system for distillation products provided by this invention demonstrates significant advantages in terms of real-time performance, adaptability, robustness, and stability. It effectively solves the inherent defects of traditional detection methods and provides reliable technical support for the intelligent control and optimization of the distillation process.

[0098] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An online detection system for the concentration of distillation products, characterized in that, The system includes a data acquisition module, which is connected to a data processing and fusion module. The data processing and fusion module is connected to a digital twin model module, which is connected to a model adaptive correction module. The model adaptive correction module is connected to a working condition pattern recognition module, which is connected to a model migration and loading module. The model migration and loading module is connected to a multi-working condition model library management module, which is connected to a sensor data module. The sensor data module is connected to a key sensor fault diagnosis module, which is connected to a fault data reconstruction module. The fault data reconstruction module is connected to a system health module. The digital twin model module is also connected to a prediction uncertainty quantification module, which is connected to a digital twin management module. The digital twin management module is connected to a closed-loop control interface module, which is connected to an edge-cloud collaborative computing module, and the edge-cloud collaborative computing module is connected to a dynamic performance evaluation module.

2. The online concentration detection system for distillation products according to claim 1, characterized in that, The data acquisition module is used to acquire real-time dynamic process data of the distillation process and discontinuous direct concentration data from at least one physical analysis instrument. The data processing and fusion module is used to preprocess and time-align the real-time dynamic process data and the discontinuous direct concentration data to form a multimodal dataset.

3. The online concentration detection system for distillation products according to claim 2, characterized in that, The digital twin model module is built based on the distillation process mechanism, receives real-time dynamic process data from the multimodal dataset as input, and outputs real-time product concentration estimates. The model adaptive correction module compares the discontinuous direct concentration data with the predicted values ​​of the digital twin model at the corresponding historical time, and dynamically adjusts the internal parameters of the digital twin model based on the comparison results.

4. The online concentration detection system for distillation products according to claim 3, characterized in that, The operating condition mode recognition module is used to automatically identify and mark the current operating condition of the distillation column based on the feed components, product specifications and operating parameters, and package and store the stable operating data under the operating condition and the corresponding calibrated digital twin model parameters as a specific operating condition model package. The mode migration loading module automatically matches the most similar working condition model package from the historical working condition model library when the system detects or receives a working condition switching command, and loads the model parameters in it as initial values ​​into the digital twin model module to achieve rapid startup and convergence under the new working condition.

5. The online concentration detection system for distillation products according to claim 4, characterized in that, The multi-condition model library management module is used to store, index, and manage multiple specific condition model packages, and record the usage frequency and prediction accuracy of each model package to provide priority reference for model matching. The sensor data module learns the causal relationships between multiple process variables in the distillation column based on historical data, and constructs a causal graph between the variables to identify abnormal data sources.

6. The online concentration detection system for distillation products according to claim 5, characterized in that, The critical sensor fault diagnosis module receives the output of the sensor data causal inference network module. When a critical sensor data changes abruptly or deviates continuously, it analyzes the logical consistency of the data with other related variables through a causal graph to determine whether the sensor has malfunctioned and triggers an alarm.

7. The online concentration detection system for distillation products according to claim 6, characterized in that, When the critical sensor fault diagnosis module confirms a sensor fault, the fault data reconstruction module uses the sensor data causal inference network module to estimate and reconstruct alternative data values ​​of the faulty sensor in real time based on the data of other normally operating sensors.

8. The online concentration detection system for distillation products according to claim 7, characterized in that, The system health module is used to assess the health status of each component of the system; when a sensor fault is diagnosed and data reconstruction is enabled, the system is automatically marked as "degraded operation" mode, and the uncertainty range of the concentration estimate at this time is assessed and reported.

9. The online concentration detection system for distillation products according to claim 8, characterized in that, The prediction uncertainty quantification module, which is connected to the digital twin model module, not only outputs the point estimate of the concentration, but also the confidence interval of the estimate, providing a risk basis for advanced control. The digital twin management module records the parameter versions after each adaptive correction of the model. When the performance of the corrected model deteriorates after evaluation, it supports one-click rollback to the previous stable version to ensure system reliability.

10. The online concentration detection system for distillation products according to claim 9, characterized in that, The closed-loop control interface module sends the final product concentration estimate and uncertainty information to the advanced process control or real-time optimization system of the distillation column in real time using a standard communication protocol, in order to form a closed-loop control loop for product quality. The edge-cloud collaborative computing module deploys data preprocessing and real-time digital twin computing on the edge to ensure real-time performance, while deploying computationally intensive tasks such as model training and operational model library management on the cloud platform to achieve optimized allocation of computing resources. The dynamic performance evaluation module automatically generates system performance reports periodically, compares the long-term deviations between soft measurement values ​​and physical sensor values, evaluates the model's accuracy and stability, and provides maintenance suggestions.

Images