Adaptive Control Simulation System Based on Digital Twin Model

Abstract

This invention discloses an adaptive control simulation system based on a digital twin model, relating to the field of gas turbine control technology. By rigorously screening the structured data stream of the gas turbine at multiple levels and dimensions, this invention significantly improves the data quality used to construct the digital twin model while ensuring data integrity. This results in a digital twin model that more closely reflects the dynamic characteristics of the real system. After embedding an adaptive control algorithm, this model can generate more accurate control parameter optimization trajectories, improving the convergence speed and stability of the simulation process. The final output control simulation strategy possesses stronger adaptability to operating conditions and dynamic response capabilities. The converted control commands can significantly improve the adjustment effect of the field controller on the gas turbine system, and effectively enhance the accuracy, robustness, and engineering practicality of the adaptive control system driven by the gas turbine digital twin.

Description

Technical Field

[0001] This invention relates to the field of gas turbine control technology, and in particular to an adaptive control simulation system based on a digital twin model. Background Technology

[0002] In the current field of industrial automation and energy power system control, gas turbines, as efficient and clean energy conversion devices, are widely used in power generation, aviation propulsion, and distributed energy systems. Their operating conditions are complex and variable, placing extremely high demands on the real-time performance, stability, and adaptability of the control system. Digital twin technology, as a core means to achieve full lifecycle monitoring and intelligent control of gas turbines, supports operating condition simulation, fault prediction, and control optimization by constructing high-fidelity virtual models. However, existing data processing methods for constructing digital twin models of gas turbines have significant drawbacks: they typically input heterogeneous operating data from multiple sources, such as temperature, pressure, speed, vibration, and fuel flow, into the model without effective data filtering. These data often contain a large number of redundant signals, low-sensitivity parameters, and abnormal information entropy data severely affected by noise interference. This results in a bloated digital twin model structure, high computational load, and difficulty in accurately reflecting key physical processes such as gas turbine combustion chamber dynamics, compressor surge boundaries, and thermal stress evolution, severely impacting simulation accuracy and real-time performance. While existing methods have attempted dimensionality reduction using correlation analysis or principal component analysis, they often rely on fixed thresholds or empirical weights, failing to adapt to the dynamic changes in gas turbine characteristics under complex operating conditions such as start-up, shutdown, variable load, and environmental disturbances. Furthermore, they lack a comprehensive assessment of data control sensitivity and information entropy characteristics, making it difficult to identify key data streams that truly characterize the system's core dynamics. In addition, existing screening mechanisms typically operate from a single dimension, failing to achieve class-level and group-level collaborative screening after data type classification, resulting in insufficient modeling ability to depict the nonlinear and strongly coupled characteristics of gas turbines. Therefore, accurately identifying and retaining highly relevant and important data from the massive structured data streams of gas turbines, while eliminating redundant and low-value information, has become a key bottleneck in improving the fidelity and control applicability of digital twin models. Without addressing these issues, digital twin models will struggle to support high-precision simulation and adaptive optimization, leading to lag in control parameter adjustments, deviations in simulation trajectories from actual operating conditions, and ultimately impacting the operational safety, regulation response speed, and energy utilization efficiency of the gas turbine.

[0003] Therefore, there is an urgent need for technical solutions for adaptive control simulation systems based on digital twin models. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an adaptive control simulation system based on a digital twin model, specifically comprising the following modules:

[0005] Data acquisition and preprocessing module: used to acquire real-time operating data of the controlled system and preprocess the operating data to obtain a structured data stream;

[0006] Digital twin model building module: Connected to the data acquisition and preprocessing module, it is used to filter structured data streams to obtain an optimized set of structured data streams, and to build a digital twin model using the optimized set of structured data streams;

[0007] Data classification and grouping unit: used to classify structured data streams according to data type and group each type of structured data stream;

[0008] Correlation analysis unit: used to analyze the correlation between each group of structured data streams in each type of structured data stream, and obtain the correlation coefficient between each group of structured data streams in each type of structured data stream;

[0009] Numerical sequence acquisition subunit: used to acquire the numerical sequences of any two sets of structured data streams in each type of structured data stream within the same time interval, denoted as the first numerical sequence and the second numerical sequence, wherein the first numerical sequence contains T sampling points arranged in chronological order, and the second numerical sequence contains T sampling points at the corresponding time points;

[0010] Numerical sequence calculation subunit: used to calculate the arithmetic mean of the first numerical sequence, which is calculated by dividing the sum of the values ​​of all sample points in the sequence by the total number of sample points T; and to calculate the arithmetic mean of the second numerical sequence, which is calculated by dividing the sum of the values ​​of all sample points in the sequence by the total number of sample points T.

[0011] Covariance calculation subunit: used to calculate the covariance between the first numerical sequence and the second numerical sequence. The calculation method is as follows: for each time point t, calculate the product of the difference between the first value and the arithmetic mean of the first numerical sequence and the difference between the second value and the arithmetic mean of the second numerical sequence at that time point, and sum the products of all time points and divide by the total number of sampling points T.

[0012] The standard deviation calculation subunit for the numerical sequence is used to calculate the standard deviation of the first numerical sequence. The calculation method is as follows: for each time point t, calculate the square of the difference between the first value at that point and the arithmetic mean of the first numerical sequence, sum the results, divide by T, and then take the square root. The standard deviation of the second numerical sequence is calculated as follows: for each time point t, calculate the square of the difference between the second value at that point and the arithmetic mean of the second numerical sequence, sum the results, divide by T, and then take the square root.

[0013] Pearson correlation coefficient calculation subunit: This unit divides the covariance obtained from the covariance calculation subunit by the product of the two standard deviations obtained from the numerical sequence standard deviation calculation subunit to obtain the Pearson correlation coefficient between the two sets of structured data streams.

[0014] The correlation coefficient calculation subunit is used to perform the numerical sequence acquisition subunit to the Pearson correlation coefficient calculation subunit on each pair of all data groups in the class to obtain the correlation coefficient between each group of structured data streams in each class of structured data streams;

[0015] The comprehensive correlation coefficient acquisition unit is used to synthesize the correlation coefficients between each group of structured data streams in each type of structured data stream to obtain the comprehensive correlation coefficient of each type of structured data stream.

[0016] Importance Analysis Unit: Used to analyze the importance of each group of structured data streams in each type of structured data stream, and obtain the importance score of each group of structured data streams in each type of structured data stream;

[0017] Output response change acquisition subunit: used to apply a disturbance signal of preset amplitude to any set of structured data streams in each type of structured data stream, and acquire the output response change of the controlled system during the disturbance period;

[0018] Control sensitivity value acquisition subunit: used to calculate the ratio of the change in the output response to the amplitude of the disturbance signal, and obtain the control sensitivity value of the structured data stream of the current group;

[0019] Group control sensitivity value acquisition subunit: used to execute the output response change acquisition subunit to the control sensitivity value acquisition subunit for each group of structured data streams to obtain the control sensitivity value of each group of structured data streams;

[0020] Information entropy value acquisition sub-unit: used to acquire time series data of each group of structured data streams, divide the time series data of each group of structured data streams into N consecutive subsequences, calculate the information entropy of each subsequence, and then calculate the average of the information entropy of all subsequences to obtain the information entropy value of the current group of structured data streams;

[0021] Information entropy score acquisition subunit: used to perform range normalization processing on the control sensitivity value and information entropy value to obtain the normalized control sensitivity score and the normalized information entropy score;

[0022] Importance score acquisition subunit: used to take the arithmetic mean of the normalized control sensitivity score and information entropy score to obtain the importance score of the structured data stream of the current group. The information entropy score acquisition subunit is executed sequentially for each group of structured data streams to obtain the importance score of each group of structured data streams in each type of structured data stream.

[0023] Comprehensive Importance Score Acquisition Unit: Used to synthesize the importance scores of each group of structured data streams in each type of structured data stream to obtain the comprehensive importance score of each type of structured data stream;

[0024] The comprehensive index acquisition unit is used to normalize the comprehensive correlation coefficient and comprehensive importance score of each type of structured data stream, and combine the normalization results to obtain the comprehensive index of each type of structured data stream.

[0025] Screening and Judgment Unit: Used to screen each group of structured data streams in each type of structured data stream by using the comprehensive correlation coefficient, comprehensive importance score, comprehensive index, correlation coefficient between each group of structured data streams in each type of structured data stream, and importance score of each group of structured data streams in each type of structured data stream, to obtain the optimized structured data streams in each type of structured data stream.

[0026] Threshold setting subunit: used to set the first threshold, second threshold, third threshold, fourth threshold and fifth threshold for the comprehensive index, comprehensive correlation coefficient, comprehensive importance score, correlation coefficient between each group of structured data streams in each type of structured data stream, and importance score of each group of structured data streams in each type of structured data stream, respectively.

[0027] The first threshold calculation interface is used to perform range normalization on the comprehensive correlation coefficient and comprehensive importance score of the structured data stream of the current category, respectively, to obtain the normalized comprehensive correlation coefficient and the normalized comprehensive importance score. The ratio of the product of the two to the arithmetic mean of the two is then used as the coupling strength factor. The 75th percentile of the coupling strength factor under normal operating conditions in the historical operating cycle is used as the coupling strength benchmark value. If the current coupling strength factor is greater than the coupling strength benchmark value, the smaller value between the normalized comprehensive correlation coefficient and the normalized comprehensive importance score is used as the first threshold; otherwise, the geometric mean of the two is used as the first threshold.

[0028] The second threshold calculation interface is used to obtain the correlation coefficients between all groups in the current category and calculate the skewness and kurtosis of their probability distribution. The 90th percentile of the skewness of the same data stream in the historical running period is used as the skewness threshold, and the 90th percentile of the kurtosis is used as the kurtosis threshold. If the current absolute value of skewness is greater than the skewness threshold and the current kurtosis is greater than the kurtosis threshold, the maximum correlation coefficient is multiplied by 0.8 as the second threshold; otherwise, the median of the correlation coefficient is multiplied by 1.1 as the second threshold.

[0029] The third threshold calculation interface is used to obtain the importance scores of all groups in the current category and calculate their coefficient of variation. The median of the coefficient of variation of the same type of structured data stream in the historical running period is used as the coefficient of variation threshold. If the current coefficient of variation is greater than the threshold, the lower quartile of the importance score is used as the third threshold. Otherwise, the arithmetic mean of the importance scores minus twice the standard deviation is used as the third threshold.

[0030] The fourth threshold calculation interface is used to obtain the correlation coefficients between all groups in each type of structured data stream within the most recent N historical running cycles, construct the first sliding time window dataset, calculate the mean and standard deviation of the first sliding time window dataset, and determine the lower limit of the 95% confidence interval based on the t-distribution, and use this lower limit as the fourth threshold.

[0031] The fifth threshold calculation interface is used to obtain the importance scores of all groups in each type of structured data stream within the most recent N historical running cycles, construct the second sliding time window dataset, calculate the mean and standard deviation of the second sliding time window dataset, and determine the lower limit of the 95% confidence interval based on the t-distribution, and use this lower limit as the fifth threshold.

[0032] The comprehensive index determination subunit is used to determine the comprehensive correlation coefficient and comprehensive importance score if the comprehensive index of the current category of structured data stream is less than the first threshold; if the comprehensive index of the current category of structured data stream is greater than or equal to the first threshold, it will proceed to the data reassembly subunit.

[0033] The comprehensive correlation coefficient and comprehensive importance score determination subunit is used to determine whether the current category's structured data stream has a comprehensive correlation coefficient less than the second threshold or a comprehensive importance score less than the third threshold. If the current category's structured data stream has a comprehensive correlation coefficient greater than or equal to the second threshold and a comprehensive importance score greater than or equal to the third threshold, it will proceed to the data reassembly subunit.

[0034] The correlation coefficient and importance score determination subunit is used to remove structured data streams of the current group from the current category's structured data stream if the correlation coefficient between structured data streams of the current category and the current group is less than the fourth threshold or the importance score of structured data streams of the current group from the current category's structured data stream is less than the fifth threshold; if the correlation coefficient between structured data streams of the current category and the current group is greater than or equal to the fourth threshold and the importance score of structured data streams of the current group from the current category's structured data stream is greater than or equal to the fifth threshold, then the data reassembly subunit is entered.

[0035] Data Reassembly Subunit: Used to reassemble the group of structured data streams that were not removed, to obtain optimized structured data streams for each type;

[0036] Digital twin model building unit: used to aggregate each type of optimized structured data stream to obtain an optimized set of structured data streams, and to build a digital twin model using the optimized set of structured data streams;

[0037] Closed-loop simulation system forming module: connected to the digital twin model building module, used to call pre-stored adaptive control algorithms and embed them into the digital twin model to form a closed-loop simulation system;

[0038] Control execution module: Connected to the closed-loop simulation system, it is used to run the simulation in the closed-loop simulation system with the current operating condition as the initial condition, generate the optimized trajectory of control parameters, generate the control simulation strategy based on the optimized trajectory of control parameters, convert the control simulation strategy into control commands, and send them to the field controller to perform adjustment operations on the controlled system.

[0039] The embodiments of the present invention have the following technical effects:

[0040] This invention significantly improves the quality of data used to build digital twin models by rigorously screening structured data streams from gas turbines at multiple levels and dimensions, while ensuring data integrity. First, it achieves structured data organization through classification and grouping. Then, it quantifies data correlation and control value by combining inter-group correlation coefficients and importance scores. Finally, it achieves refined group-level screening by dynamically generating first to fifth thresholds using a combination of comprehensive correlation coefficients, comprehensive importance scores, and comprehensive indices. This effectively eliminates redundant, low-relevance, low-sensitivity, and information entropy-abnormal data groups, retaining core data that significantly characterize key processes such as gas turbine combustion dynamics, compressor characteristics, and thermal stress evolution. Based on the data flow, a digital twin model that more closely resembles the dynamic characteristics of the real system is constructed. After embedding an adaptive control algorithm, this model can generate a more accurate control parameter optimization trajectory, improve the convergence speed and stability of the simulation process, and the final output control simulation strategy has stronger adaptability to operating conditions and dynamic response capabilities. The converted control commands can significantly improve the regulation effect of the field controller on the gas turbine system, reduce combustion oscillation, suppress the risk of compressor surge, shorten the response time of variable load, and improve thermal efficiency. Overall, it realizes closed-loop optimization from data acquisition to control execution, and improves the accuracy, robustness, and engineering practicality of the adaptive control system driven by the gas turbine digital twin. Attached Figure Description

[0041] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0042] Figure 1 This is a framework diagram of an adaptive control simulation system based on a digital twin model provided in an embodiment of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0044] Example 1: As Figure 1 As shown, the adaptive control simulation system based on a digital twin model provided by this invention includes the following modules:

[0045] Data acquisition and preprocessing module: used to acquire real-time operating data of the controlled system and preprocess the operating data to obtain a structured data stream;

[0046] It is worth noting that real-time operating data is collected through a sensor network deployed in the gas turbine body and auxiliary systems. This sensor network includes temperature sensors, pressure sensors, vibration sensors, speed sensors, flow sensors, and actuator position feedback sensors. The collected data covers key operating parameters such as combustion chamber outlet temperature field distribution, compressor inlet and outlet pressure and temperature, turbine interstage temperature, fuel supply flow and pressure, air intake flow, adjustable guide vane opening, bearing vibration acceleration signal, rotor speed, and exhaust temperature. The sampling frequency is set differently according to the dynamic characteristics of the parameters. The sampling period for temperature, pressure, and speed signals is 10ms, and the sampling frequency for vibration signals is 5kHz. All data is transmitted to the edge computing unit via industrial Ethernet. The raw operating data is preprocessed to generate a structured data stream. The preprocessing process includes: first... First, data cleaning is performed to identify and remove outliers and missing data points caused by sensor malfunctions or communication interruptions. Outliers are detected using a statistical filtering method based on the 3σ criterion, and data is repaired using linear interpolation or a mean compensation method based on a sliding window. Second, time synchronization processing is performed, using the rotational speed pulse signal as the reference clock source to align the time of multi-channel asynchronous sampling data, ensuring that all parameters have physical consistency at the same timestamp. Subsequently, unit normalization and dimension unification are performed, converting different physical quantities into standardized engineering units, and range normalization is performed on variables with large differences in numerical range. Finally, the processed data is organized into a triplet structure of "parameter type-measuring point location-time series" and encapsulated into a timestamp sequence data package with a unified data format, forming a structured data stream with clear structure and semantics for use by subsequent classification and filtering modules.

[0047] Digital twin model building module: Connected to the data acquisition and preprocessing module, it is used to filter structured data streams to obtain an optimized set of structured data streams, and to build a digital twin model using the optimized set of structured data streams;

[0048] It is worth noting that the process of constructing a digital twin model of a gas turbine using the optimized structured data stream set obtained through screening includes three stages: physical modeling, parameter identification, and model calibration. First, in the physical modeling stage, mechanistic models of core components are established based on the aerodynamic and thermodynamic principles of the gas turbine, including a compressor characteristic mapping model, a combustion chamber energy release dynamic model, a turbine work model, and a rotor dynamics model. Each sub-model is coupled and connected through mass conservation, energy conservation, and momentum conservation equations, forming a basic simulation framework reflecting the overall dynamic characteristics of the unit. Second, in the parameter identification stage, the optimized structured data stream is input into the corresponding sub-models according to four categories: thermodynamic, aerodynamic, mechanical, and control. The least squares method and the recursive augmented least squares algorithm are used to perform online calibration of key unknown parameters. Estimates, such as compressor efficiency correction coefficients, combustion chamber volume time constants, and bearing damping coefficients, are made to ensure that model parameters are consistent with actual equipment conditions. Finally, in the model calibration phase, a multi-objective optimization strategy is adopted, using measured exhaust temperature, speed response, and adjustable guide vane trajectory as reference outputs. The internal coupling weights and delay elements of the model are adjusted to ensure that the dynamic response error between the simulation output and the real system under typical operating conditions (such as startup, full-load operation, and 50% step load change) is less than 3%. Residual analysis is used to verify the white noise characteristics of the model residuals, ensuring that unmodeled dynamic influences are minimized. The constructed digital twin model possesses high fidelity and timeliness, accurately reproducing key processes such as gas turbine combustion dynamics, compressor surge boundary evolution, and thermal stress accumulation. This provides a reliable simulation environment for subsequent embedding of adaptive control algorithms, supports the generation of accurate control parameter optimization trajectories, and improves the regulation quality and operational safety of the closed-loop control system.

[0049] Data classification and grouping unit: used to classify structured data streams according to data type and group each type of structured data stream;

[0050] It is worth noting that after generating the structured data stream, the data stream is first classified according to its physical attributes and functional characteristics, dividing all data into four categories: The first category is thermodynamic parameters, including temperature-related data streams such as combustion chamber temperature, turbine interstage temperature, and exhaust temperature; the second category is aerodynamic parameters, including pressure and flow-related data streams such as compressor inlet and outlet pressure, fuel supply pressure, air intake pressure, compressor inlet and outlet flow rate, and fuel flow rate; the third category is mechanical dynamics, including data streams reflecting the mechanical operating status of the unit, such as vibration acceleration signals at various bearing positions, rotor speed, and axial displacement; and the fourth category is control and execution, including feedback signals from actuators such as adjustable guide vane opening, fuel control valve opening, and anti-surge valve operating position. Each type of structured data stream is further grouped according to the spatial distribution of measuring points and the system functional modules. Among them, the thermodynamic parameters are divided into 3 groups according to the combustion section: measuring points 1 to 4 evenly distributed around the combustion chamber are the first group, measuring points 5 to 8 are the second group, and measuring points between turbine stages and exhaust are the third group; the aerodynamic parameters are divided into 2 groups according to the air path flow: measuring points related to the compressor end are the first group, and measuring points related to fuel supply and combustion chamber intake are the second group; the mechanical dynamics are divided into 3 groups according to the rotor support structure: the front bearing vibration and speed signal are the first group, the mid-section bearing vibration signal is the second group, and the tail bearing vibration signal is the third group; the control execution is divided into 2 groups according to the control subsystem: the intake regulation related execution mechanism is the first group, and the fuel regulation related execution mechanism is the second group. The above grouping method ensures that the data streams within each group have similar dynamic response characteristics and physical correlations, while avoiding data redundancy. The structured data streams within each group are independent time series of different measurement points or different parameters, without duplicate parameters or spatially overlapping measurement points. For example, the same group does not contain temperature signals from the same location in two combustion chambers, nor does it mix speed signals and vibration signals into the same group. This ensures that the correlation analysis of the data within the group has clear physical meaning, and the boundaries between groups are clear, providing a reasonable and interpretable data organization basis for subsequent multi-dimensional screening based on inter-group correlation and importance.

[0051] Correlation analysis unit: used to analyze the correlation between each group of structured data streams in each type of structured data stream, and obtain the correlation coefficient between each group of structured data streams in each type of structured data stream;

[0052] Numerical sequence acquisition subunit: used to acquire the numerical sequences of any two sets of structured data streams in each type of structured data stream within the same time interval, denoted as the first numerical sequence and the second numerical sequence, wherein the first numerical sequence contains T sampling points arranged in chronological order, and the second numerical sequence contains T sampling points at the corresponding time points;

[0053] Numerical sequence calculation subunit: used to calculate the arithmetic mean of the first numerical sequence, which is calculated by dividing the sum of the values ​​of all sample points in the sequence by the total number of sample points T; and to calculate the arithmetic mean of the second numerical sequence, which is calculated by dividing the sum of the values ​​of all sample points in the sequence by the total number of sample points T.

[0054] Covariance calculation subunit: used to calculate the covariance between the first numerical sequence and the second numerical sequence. The calculation method is as follows: for each time point t, calculate the product of the difference between the first value and the arithmetic mean of the first numerical sequence and the difference between the second value and the arithmetic mean of the second numerical sequence at that time point, and sum the products of all time points and divide by the total number of sampling points T.

[0055] The standard deviation calculation subunit for the numerical sequence is used to calculate the standard deviation of the first numerical sequence. The calculation method is as follows: for each time point t, calculate the square of the difference between the first value at that point and the arithmetic mean of the first numerical sequence, sum the results, divide by T, and then take the square root. The standard deviation of the second numerical sequence is calculated as follows: for each time point t, calculate the square of the difference between the second value at that point and the arithmetic mean of the second numerical sequence, sum the results, divide by T, and then take the square root.

[0056] Pearson correlation coefficient calculation subunit: This unit divides the covariance obtained from the covariance calculation subunit by the product of the two standard deviations obtained from the numerical sequence standard deviation calculation subunit to obtain the Pearson correlation coefficient between the two sets of structured data streams.

[0057] The correlation coefficient calculation subunit is used to perform the numerical sequence acquisition subunit to the Pearson correlation coefficient calculation subunit on each pair of all data groups in the class to obtain the correlation coefficient between each group of structured data streams in each class of structured data streams;

[0058] It is worth noting that the numerical sequence acquisition subunit to the correlation coefficient calculation subunit has the following technical effects: by calculating the Pearson correlation coefficient of any two sets of data sequences in various structured data streams of gas turbines, it accurately quantifies their linear correlation degree, providing a reliable basis for the subsequent generation of comprehensive correlation coefficients. It ensures that highly correlated redundant data groups can be accurately identified and removed during the screening process, avoiding invalid data from interfering with the construction of the digital twin model, improving the model's ability to represent the core dynamic processes of gas turbines, and thus enhancing the accuracy and response capability of the control simulation strategy.

[0059] The comprehensive correlation coefficient acquisition unit is used to synthesize the correlation coefficients between each group of structured data streams in each type of structured data stream to obtain the comprehensive correlation coefficient of each type of structured data stream.

[0060] It is worth noting that after calculating the correlation coefficients between all groups in each type of structured data stream, the arithmetic mean method is used to synthesize all the correlation coefficients within the same category to obtain the comprehensive correlation coefficient for that type of structured data stream. Specifically, the correlation coefficients between all groups belonging to the same category are grouped into a numerical set. For example, if the thermodynamic parameter class contains 3 sets of data, there are 3 pairwise correlation coefficients (group 1-group 2, group 1-group 3, group 2-group 3). These values ​​are added together and then divided by the total number of pairs; the comprehensive correlation coefficient is calculated as follows: For other categories, the same logic applies. If a category contains M data groups, the total number of correlation coefficients between groups is C(M,2) = M(M-1) / 2. The overall correlation coefficient is... ,in This represents the Pearson correlation coefficient between group i and group j. This method obtains a single index characterizing the overall correlation level of the data in a class by averaging the correlation strength among all groups within the class. It reflects the tightness of coupling within the parameter structure of that class, providing a quantitative basis for subsequent class-level selection.

[0061] Importance Analysis Unit: Used to analyze the importance of each group of structured data streams in each type of structured data stream, and obtain the importance score of each group of structured data streams in each type of structured data stream;

[0062] Output response change acquisition subunit: used to apply a disturbance signal of preset amplitude to any set of structured data streams in each type of structured data stream, and acquire the output response change of the controlled system during the disturbance period;

[0063] It is worth noting that a perturbation signal with a preset amplitude is applied to any set of structured data streams in each type of structured data stream. The "preset amplitude" is determined based on the parameter type, operating safety boundary, and historical variable operating condition data statistics. The specific setting principle is to excite a measurable system response while ensuring the safe operation of the gas turbine. For control input signals such as fuel flow and adjustable guide vane opening, the perturbation amplitude is set to ±5% of the current steady-state value. For example, if the fuel reference flow rate is 8.2 kg / s under rated load, the perturbation signal is a step change of ±0.41 kg / s. For process feedback signals such as temperature and pressure, since they are controlled variables rather than directly adjustable variables, the perturbation is applied indirectly through associated actuators. For example, when perturbing the combustion chamber temperature group, a fuel bias of ±3% is introduced by adjusting the fuel distribution valve opening to simulate local combustion intensity changes. All perturbation signals are injected in the form of pseudo-random binary sequences (PRBS) for a duration of 30 seconds to ensure that the dynamic process of the system is excited while avoiding long-term deviation from the set value. During the disturbance process, key safety indicators such as vibration amplitude, exhaust temperature deviation, and compressor margin are monitored in real time. If the preset safety threshold is exceeded, the disturbance is terminated immediately and control is restored to ensure the safety of the test.

[0064] Control sensitivity value acquisition subunit: used to calculate the ratio of the change in the output response to the amplitude of the disturbance signal, and obtain the control sensitivity value of the structured data stream of the current group;

[0065] It is worth noting that the core advantage of using the ratio of the change in output response to the amplitude of the disturbance signal as the control sensitivity value lies in achieving a unified quantification, comparability, and physical interpretability of the control value of the data set. Specifically, this ratio is essentially a static gain estimate of the system at a local operating point, reflecting the intensity of the output change caused by a unit input disturbance. For example, the larger the ratio of fuel flow disturbance to speed response change, the more significant the impact of the channel containing that data set on speed regulation. This calculation method does not rely on absolute numerical values, eliminating the influence of differences in dimensions and operating conditions, allowing the importance of different parameter types (such as temperature, pressure, and opening degree) to be compared on the same scale, providing a foundation for subsequent normalization and fusion. The control sensitivity value plays a crucial role in this invention: it directly characterizes the functional weight of a set of structured data streams in the gas turbine control loop. High sensitivity means that the data is strongly correlated with core control objectives (such as power regulation, surge control, and temperature protection), and is a key information source that the digital twin model must retain. For example, if a disturbance at a combustion chamber temperature measuring point causes a significant change in turbine temperature with a rapid response, its control sensitivity value is high, indicating that the measuring point has high characterization value for thermal stress management. Conversely, if a disturbance at an auxiliary pressure measuring point causes almost no change in system output, its sensitivity is low and can be identified as redundant information. By introducing control sensitivity, this invention overcomes the limitations of traditional methods that rely solely on correlation analysis, incorporating causal response capability into the data screening criteria. This ensures that the constructed digital twin model not only reflects the statistical correlation between data but also captures the actual control action path, thereby improving the model's prediction accuracy and strategy generation quality in adaptive control simulation, ultimately enabling control commands to have stronger dynamic adjustment capabilities and safety guarantees.

[0066] Each control sensitivity value acquisition subunit: is used to execute the output response change acquisition subunit to the control sensitivity value acquisition subunit for each group of structured data streams to obtain the control sensitivity value of each group of structured data streams;

[0067] Information entropy value acquisition sub-unit: used to acquire time series data of each group of structured data streams, divide the time series data of each group of structured data streams into N consecutive subsequences, calculate the information entropy of each subsequence, and then calculate the average of the information entropy of all subsequences to obtain the information entropy value of the current group of structured data streams;

[0068] It is worth noting that for each set of structured data stream time series data, it is first divided into N consecutive subsequences of equal length according to time sequence. The subsequence length is determined based on the signal sampling frequency and dynamic characteristics. For low-frequency parameters such as temperature, pressure, and flow rate (sampling period 10ms), the subsequence length is set to 100 sampling points, i.e., each segment is 1 second long; for high-frequency signals such as vibration acceleration (sampling frequency 5kHz), the subsequence length is set to 500 sampling points, i.e., each segment is 0.1 seconds long, to balance time series resolution and statistical stability. For each subsequence, its information entropy value is calculated to characterize the uncertainty and information richness of the data within that time period. The specific calculation process is as follows: Suppose a subsequence contains M values, denoted as Mi. First, the subsequence is normalized so that all values ​​are mapped to the interval [0,1]. Then, the normalized range of values ​​is divided into K equal-width intervals (K=8), and the frequency of values ​​in each interval is counted to obtain the probability distribution. ,in This represents the frequency of the i-th interval; finally, the information entropy value H of the subsequence is calculated using the Shannon information entropy formula, expressed as: i ranges from 1 to K, when When the value is 0, the corresponding item takes the value of 0. After calculating the information entropy of all N subsequences, the arithmetic mean is taken as the information entropy value of the current group's structured data stream. The higher the value, the more complex the data fluctuations, the greater the amount of information contained, and the better it can reflect the subtle changes in the gas turbine's operating status, providing a quantitative basis for the comprehensive judgment of the subsequent importance score.

[0069] Information entropy score acquisition subunit: used to perform range normalization processing on the control sensitivity value and information entropy value to obtain the normalized control sensitivity score and the normalized information entropy score;

[0070] Importance score acquisition subunit: used to take the arithmetic mean of the normalized control sensitivity score and information entropy score to obtain the importance score of the structured data stream of the current group. The information entropy score acquisition subunit is executed sequentially for each group of structured data streams to obtain the importance score of each group of structured data streams in each type of structured data stream.

[0071] It is worth noting that the sub-unit for obtaining output response change has the following technical effects: it obtains the output response change by applying a disturbance signal and calculates the control sensitivity value. Combined with information entropy analysis, it quantifies the degree of influence and information richness of each set of structured data streams on the control behavior of the gas turbine system. After normalization, the data is merged into an importance score. This makes the screening process not only consider the data correlation, but also take into account its actual value in control regulation. It effectively retains data sets with high influence on key control objectives such as combustion stability and compressor condition regulation, thereby improving the control guidance and simulation optimization capabilities of the digital twin model.

[0072] Comprehensive Importance Score Acquisition Unit: Used to synthesize the importance scores of each group of structured data streams in each type of structured data stream to obtain the comprehensive importance score of each type of structured data stream;

[0073] It is worth noting that after obtaining the importance score of each group of structured data streams within each category, the arithmetic mean method is used to synthesize the importance scores of all groups within the same category to obtain the overall importance score of that category of structured data streams. Specifically, the process involves summing the importance scores of each group within the category and then dividing by the number of groups. For example, the thermodynamic parameter category is divided into 3 groups, with importance scores of... Then the overall importance score = If a category contains M groups, then the overall importance score = ,in This represents the importance score of group i. This method obtains a comprehensive index reflecting the overall control value of the data by averaging and fusing the control sensitivity and information entropy characteristics of each group within a class. This index reflects the overall impact of the parameters on the core operational objectives of the gas turbine (such as combustion stability, thermal efficiency, and safety margin).

[0074] The comprehensive index acquisition unit is used to normalize the comprehensive correlation coefficient and comprehensive importance score of each type of structured data stream, and combine the normalization results to obtain the comprehensive index of each type of structured data stream.

[0075] It is worth noting that after normalizing the comprehensive correlation coefficient and comprehensive importance score of each type of structured data stream, an arithmetic mean method is used to combine them to obtain the comprehensive index of that type of structured data stream. The specific process is as follows: First, the comprehensive correlation coefficient and comprehensive importance score are respectively normalized to map them to the [0,1] interval. The normalization formula is: x'=(x-x_min) / (x_max-x_min), where x is the original value, and x_min and x_max are the minimum and maximum values ​​of this indicator among all current categories, respectively. Then, the arithmetic mean of the normalized comprehensive correlation coefficient c' and the normalized comprehensive importance score s' is taken, i.e., the comprehensive index = (c'+s') / 2. This comprehensive index reflects the overall performance of a type of data in both internal correlation and control value dimensions. As the core criterion for category-level screening, it is used to initially identify and eliminate data categories with loose correlation and low control value, ensuring that the data streams entering the subsequent fine-tuning stage have the potential to become effective inputs for the digital twin model.

[0076] Screening and Judgment Unit: Used to screen each group of structured data streams in each type of structured data stream by using the comprehensive correlation coefficient, comprehensive importance score, comprehensive index, correlation coefficient between each group of structured data streams in each type of structured data stream, and importance score of each group of structured data streams in each type of structured data stream, to obtain the optimized structured data streams in each type of structured data stream.

[0077] Threshold setting subunit: used to set the first threshold, second threshold, third threshold, fourth threshold and fifth threshold for the comprehensive index, comprehensive correlation coefficient, comprehensive importance score, correlation coefficient between each group of structured data streams in each type of structured data stream, and importance score of each group of structured data streams in each type of structured data stream, respectively.

[0078] The first threshold calculation interface is used to perform range normalization on the comprehensive correlation coefficient and comprehensive importance score of the structured data stream of the current category, respectively, to obtain the normalized comprehensive correlation coefficient and the normalized comprehensive importance score. The ratio of the product of the two to the arithmetic mean of the two is then used as the coupling strength factor. The 75th percentile of the coupling strength factor under normal operating conditions in the historical operating cycle is used as the coupling strength benchmark value. If the current coupling strength factor is greater than the coupling strength benchmark value, the smaller value between the normalized comprehensive correlation coefficient and the normalized comprehensive importance score is used as the first threshold; otherwise, the geometric mean of the two is used as the first threshold.

[0079] It is worth noting that the coupling strength factor plays a core role in guiding decision-making in this invention. It quantifies the degree of synergy between the two by calculating the ratio of the product of the normalized comprehensive correlation coefficient and the comprehensive importance score to its arithmetic mean. When this ratio is greater than the 75th percentile under historical normal conditions, it indicates that high correlation and high importance are highly consistent in the current category, and the overall data quality is excellent. In this case, the smaller of the two values ​​is used as the first threshold, reflecting the principle of "weakest link control" to ensure that any dimension that fails to meet the standard is strictly screened. Conversely, it indicates that there is a disconnect between correlation and importance, and there may be a phenomenon of "high correlation but low control sensitivity" or "high control sensitivity but weak correlation". In this case, the geometric mean is used as the first threshold to balance the influence of the two indicators and avoid a single indicator dominating the decision. This mechanism enables threshold generation to be adaptive under operating conditions, accurately responding to the drift of data characteristics during dynamic processes such as gas turbine start-up and shutdown and load changes. It ensures that the screening results always focus on the core data stream that has a comprehensive characterization capability for key control objectives such as combustion stability, compressor surge prevention, and thermal stress management, thereby constructing a high-fidelity digital twin model and improving the accuracy of control simulation strategies and the effectiveness of on-site adjustment.

[0080] The second threshold calculation interface is used to obtain the correlation coefficients between all groups in the current category and calculate the skewness and kurtosis of their probability distribution. The 90th percentile of the skewness of the same data stream in the historical running period is used as the skewness threshold, and the 90th percentile of the kurtosis is used as the kurtosis threshold. If the current absolute value of skewness is greater than the skewness threshold and the current kurtosis is greater than the kurtosis threshold, the maximum correlation coefficient is multiplied by 0.8 as the second threshold; otherwise, the median of the correlation coefficient is multiplied by 1.1 as the second threshold.

[0081] It is worth noting that the calculation methods for skewness and kurtosis are as follows: Obtain the dataset consisting of the correlation coefficients between all groups in the current category, and calculate the skewness and kurtosis of its probability distribution to characterize the distribution shape. Skewness reflects the asymmetry of the data distribution and is calculated as the ratio of the third central moment to the cube of the standard deviation, expressed as: Skewness = ,in For the i-th correlation coefficient value, Let σ be the mean, σ be the standard deviation, and M be the total number of correlation coefficients between groups. A large absolute value of skewness indicates that most correlation coefficients are concentrated in high or low value regions, suggesting a significant strong or weak correlation dominance. Kurtosis reflects the thickness of the distribution's tails and is calculated as the ratio of the fourth central moment to the square of the variance minus 3 (i.e., excess kurtosis). The expression is: Kurtosis = When kurtosis is greater than 0, it indicates that the distribution has a peak and thick tail, with a few extremely high correlation coefficient values, corresponding to redundant sensors or strong coupling interference paths. In this invention, skewness and kurtosis are used together to identify abnormal patterns in the correlation structure of the current category of data: if the absolute value of skewness is greater than the 90th percentile of historical data of the same type and kurtosis is greater than the threshold, it indicates that there are a few extremely highly correlated groups in this category of data while the correlation of the remaining groups is generally low, which is very likely a redundant structure caused by repeated measurement points or signal crosstalk. At this time, the maximum correlation coefficient is multiplied by 0.8 as the second threshold to actively lower the screening threshold and ensure that such strongly correlated groups are included in the subsequent judgment process, avoiding the accidental deletion of key coupling relationships; otherwise, the median is multiplied by 1.1 as the second threshold to appropriately relax the standard to retain potentially valuable groups. This mechanism enables the correlation screening to have distribution perception capabilities, and can adaptively identify changes in the data association structure of gas turbines under different operating conditions, improving the representation accuracy of digital twin models for complex dynamic processes such as combustion oscillation propagation paths and compressor stall precursors.

[0082] The third threshold calculation interface is used to obtain the importance scores of all groups in the current category and calculate their coefficient of variation. The median of the coefficient of variation of the same type of structured data stream in the historical running period is used as the coefficient of variation threshold. If the current coefficient of variation is greater than the threshold, the lower quartile of the importance score is used as the third threshold. Otherwise, the arithmetic mean of the importance scores minus twice the standard deviation is used as the third threshold.

[0083] It is worth noting that the calculation method for the coefficient of variation is as follows: Obtain the importance scores of all groups within the current category, and calculate their coefficient of variation to measure the degree of data dispersion. The coefficient of variation is defined as the ratio of the standard deviation to the mean, expressed as: CV = σ / μ, where σ is the standard deviation of the importance scores and μ is their arithmetic mean. This indicator eliminates the influence of dimensions and is suitable for comparing volatility between different categories. In this invention, the coefficient of variation is used to determine the degree of difference in control value among data groups within the current category: if it is greater than the median of historical data of the same type, it indicates that there are groups in this category with control sensitivity significantly higher or lower than the average level, which includes core parameters (such as combustion chamber temperature feedback) that play a crucial role in the safe operation of the gas turbine or completely irrelevant auxiliary signals. In this case, the lower quartile is used as the third threshold to strictly screen and retain only high-importance groups, preventing low-value data from interfering with model construction; conversely, if it is lower than the median, it indicates that the importance distribution of each group is relatively uniform, and the mean minus twice the standard deviation is used as the third threshold to ensure screening strength while avoiding excessive rejection. This strategy ensures that importance screening can both identify prominent key variables and accommodate normal fluctuations, thereby improving the model's ability to capture real control-sensitive paths.

[0084] The fourth threshold calculation interface is used to obtain the correlation coefficients between all groups in each type of structured data stream within the most recent N historical running cycles, construct the first sliding time window dataset, calculate the mean and standard deviation of the first sliding time window dataset, and determine the lower limit of the 95% confidence interval based on the t-distribution, and use this lower limit as the fourth threshold.

[0085] The fifth threshold calculation interface is used to obtain the importance scores of all groups in each type of structured data stream within the most recent N historical running cycles, construct the second sliding time window dataset, calculate the mean and standard deviation of the second sliding time window dataset, and determine the lower limit of the 95% confidence interval based on the t-distribution, and use this lower limit as the fifth threshold.

[0086] It is worth noting that the first to fifth threshold calculation interfaces have the following technical effects: by establishing a dynamic threshold generation mechanism based on the statistical characteristics of historical operating data, they completely eliminate the problems of poor adaptability and weak generalization ability caused by traditional fixed thresholds or experience-based settings, and achieve adaptive optimization of the screening criteria: The setting of the first threshold introduces a coupling strength factor. By comparing the degree of synergy between the current comprehensive correlation and comprehensive importance, it dynamically selects the strategy of "taking the smaller value" or "taking the geometric mean," so that in a high-synergy state, it emphasizes the weakest indicators, and in a low-synergy state, it focuses on overall balance, improving the intelligence and rationality of the threshold setting; The second threshold combines the skewness and kurtosis of the current inter-group correlation coefficient distribution, and uses the 90th percentile of historical similar data as a dynamic benchmark to identify extreme correlation patterns, preventing the overall judgment from being misled by a few strongly correlated groups, and ensuring accurate identification of redundant data; The third threshold adaptively selects the lower quartile or the mean minus two times the standard based on the comparison results of the coefficient of variation and the historical median. The standard deviation is used as a criterion to effectively address data volatility and improve screening stability. The fourth and fifth thresholds are generated based on the lower limit of the t-score confidence interval of the correlation coefficient and importance score between historical groups within the sliding time window. This fully considers the uncertainty of samples and the impact of operating condition drift, allowing the thresholds to be dynamically adjusted with the operating environment and possessing strong time-series adaptability. This entire threshold generation method integrates distribution pattern analysis, statistical inference, and dynamic benchmark mechanisms. It not only avoids the subjectivity of human parameter setting but also enables the screening process to accurately respond to changes in the data characteristics of the gas turbine under different operating conditions. This ensures that the data groups to be removed are truly low-value redundant information, while the data groups to be retained have high representativeness and control sensitivity. This allows the construction of a digital twin model that closely approximates the real system, significantly improving the simulation convergence speed and the accuracy of control parameter optimization. The output control simulation strategy has excellent dynamic adjustment capabilities, and the converted control commands can significantly improve the field controller's control effect on the complex dynamic processes of the gas turbine, achieving a synergistic improvement in safety, responsiveness, and energy efficiency.

[0087] The comprehensive index determination subunit is used to determine the comprehensive correlation coefficient and comprehensive importance score if the comprehensive index of the current category of structured data stream is less than the first threshold; if the comprehensive index of the current category of structured data stream is greater than or equal to the first threshold, it will proceed to the data reassembly subunit.

[0088] The comprehensive correlation coefficient and comprehensive importance score determination subunit is used to determine whether the current category's structured data stream has a comprehensive correlation coefficient less than the second threshold or a comprehensive importance score less than the third threshold. If the current category's structured data stream has a comprehensive correlation coefficient greater than or equal to the second threshold and a comprehensive importance score greater than or equal to the third threshold, it will proceed to the data reassembly subunit.

[0089] The correlation coefficient and importance score determination subunit is used to remove structured data streams of the current group from the current category's structured data stream if the correlation coefficient between structured data streams of the current category and the current group is less than the fourth threshold or the importance score of structured data streams of the current group from the current category's structured data stream is less than the fifth threshold; if the correlation coefficient between structured data streams of the current category and the current group is greater than or equal to the fourth threshold and the importance score of structured data streams of the current group from the current category's structured data stream is greater than or equal to the fifth threshold, then the data reassembly subunit is entered.

[0090] Data Reassembly Subunit: Used to reassemble the group of structured data streams that were not removed, to obtain optimized structured data streams for each type;

[0091] It is worth noting that the threshold setting subunit to the data reassembly subunit has the following technical effects: by constructing a multi-level, multi-condition coupled screening logic framework, it achieves refined and intelligent screening of gas turbine structured data streams: First, a preliminary judgment is made based on the comprehensive index. If it is lower than the first threshold, the comprehensive correlation coefficient and comprehensive importance score of this type of data stream are further verified to see if they are both lower than their respective thresholds, in order to identify low-quality data categories with weak overall correlation and low control value. For data categories that fail the category-level screening, the process goes deeper to the group level, where the inter-group correlation coefficient and importance score of each data group within it are double-verified. Only when both indicators are lower than the corresponding thresholds are they removed, avoiding the accidental deletion of key data groups due to a single abnormal indicator, thereby ensuring data integrity. While ensuring integrity, the system effectively removes redundant, low-sensitivity, and information entropy-abnormal data sets. This hierarchical progressive screening mechanism achieves closed-loop control from "class-level coarse screening" to "group-level fine screening," significantly improving the robustness and discrimination accuracy of the screening process. It ensures that the final retained data sets can accurately characterize core physical processes such as gas turbine combustion dynamics, compressor aerodynamic characteristics, and rotor thermal stress evolution, providing high-quality input for building a high-fidelity digital twin model. This, in turn, enables the simulation system with embedded adaptive control algorithms to have stronger dynamic response capabilities and operating condition adaptability. The generated optimized control parameter trajectory is closer to actual operating requirements, and the control commands can effectively suppress combustion oscillations, prevent compressor surge, shorten the transition time of variable loads, and improve thermal efficiency, comprehensively enhancing the stability and energy efficiency of the gas turbine control system.

[0092] Digital twin model building unit: used to aggregate each type of optimized structured data stream to obtain an optimized set of structured data streams, and to build a digital twin model using the optimized set of structured data streams;

[0093] It is worth noting that the data classification and grouping unit to the digital twin model construction unit has the following technical effects: it achieves structured organization of data by classifying and grouping the structured data stream of the gas turbine, and further analyzes the correlation and importance between groups, comprehensively obtaining the comprehensive correlation coefficient, comprehensive importance score and comprehensive index of each type of data stream. Combined with multi-dimensional indicators, it performs fine screening of each group of data streams, effectively eliminating redundant, low-correlation, low-sensitivity and information entropy abnormal data groups, and retaining the core data streams that have significant characterization capabilities for key processes such as gas turbine combustion dynamics, compressor characteristics and thermal stress evolution. This allows for the construction of a digital twin model that is closer to the dynamic characteristics of the real system, laying the data foundation for subsequent generation of high-precision control parameter optimization trajectories and improvement of control command regulation effects. Ultimately, it enables precise regulation of the gas turbine system, reducing combustion oscillation, suppressing compressor surge risk, shortening variable load response time and improving thermal efficiency.

[0094] Closed-loop simulation system forming module: connected to the digital twin model building module, used to call pre-stored adaptive control algorithms and embed them into the digital twin model to form a closed-loop simulation system;

[0095] It is worth noting that the pre-stored adaptive control algorithms include Model Reference Adaptive Control (MRAC), self-tuning PID control, gain scheduling adaptive control, neural network adaptive control, and online optimization control algorithms based on reinforcement learning. This invention preferentially chooses Model Reference Adaptive Control (MRAC) as the embedded algorithm because it has a clear stability theoretical guarantee and can adjust the controller parameters in real time according to the desired dynamic characteristics of the reference model. It is particularly suitable for compensating for dynamic characteristic changes in gas turbines operating over a wide range of conditions. In practical implementation, a reference model matching the expected performance of the gas turbine is first constructed within the digital twin model, for example, by setting an ideal speed response curve or an upper limit for the exhaust temperature change rate. Then, the MRAC controller is integrated modularly into the control interface layer of the digital twin model. Its structure includes an adaptive law module and a parameter update mechanism. By comparing the simulation output of the digital twin model with the output deviation of the reference model in real time, parameter adjustment laws are designed based on Lyapunov stability theory to dynamically correct key control parameters such as fuel flow regulation gain and adjustable guide vane response rate. This controller and the digital twin model together constitute a closed-loop simulation system, where the controller output acts on the model input, and the model output feeds back to the controller to form a negative feedback loop, enabling online optimization and verification of the control strategy. This embedding method ensures deep integration between the control algorithm and the high-fidelity model, allowing the simulation process not only to predict system behavior but also to autonomously adjust control parameters to approximate optimal performance, providing a reliable foundation for subsequently generating high-quality control parameter optimization trajectories.

[0096] Control execution module: It is connected to the closed-loop simulation system to form a module. It is used to run the simulation in the closed-loop simulation system with the current working condition as the initial condition, generate the optimized trajectory of control parameters, generate the control simulation strategy based on the optimized trajectory of control parameters, convert the control simulation strategy into control commands, and send them to the field controller to perform adjustment operations on the controlled system.

[0097] It is worth noting that in the closed-loop simulation system, the simulation is started using the real-time operating state of the controlled system as the initial condition. This initial condition includes measured values ​​of key parameters such as current engine speed, combustion chamber temperature, compressor pressure ratio, fuel flow rate, and adjustable guide vane opening, ensuring a high degree of consistency between the simulation environment and the current operating point of the physical system. During the simulation, the model dynamically responds to changes in setpoints or external disturbances under the regulation of the MRAC controller, generating optimized adjustment trajectories for control variables such as fuel valve opening and guide vane angle. This trajectory represents a set of control parameter sequences that evolve over time, exhibiting characteristics of smooth transition, overshoot suppression, and rapid convergence. Based on this optimized trajectory, the variation patterns of key control nodes are extracted, generating a multi-... The control simulation strategy for staged regulation, such as in scenarios with a step change in load, includes predictive fuel feedforward, coordinated regulation of compressor guide vanes, and ramp rate control under thermal stress constraints. Subsequently, this control simulation strategy is converted into a set of control commands recognizable by the field controller through a standardized communication protocol, including analog output commands (4–20mA signals) and digital action commands (Modbus / TCP messages). After security verification, these commands are sent to the gas turbine field distributed control system (DCS) or PLC controller to drive the actuators to perform regulation operations on the controlled system. The entire process achieves a closed-loop mapping from high-fidelity simulation to actual control, significantly improving the stability and response accuracy of the regulation process.

[0098] It should be noted that the terminology used in this invention is for describing specific embodiments only and is not intended to limit the scope of this application. As shown in this specification, unless the context clearly indicates otherwise, words such as "a," "an," "an," and / or "the" do not specifically refer to the singular and may include the plural. The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element.

[0099] It should also be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Unless otherwise expressly specified and limited, the terms "installed," "connected," "linked," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. For those skilled in the art, the specific meaning of the above terms in the present invention can be understood according to the specific circumstances.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive control simulation system based on a digital twin model, characterized in that, Includes the following modules: Data acquisition and preprocessing module: used to acquire real-time operating data of the controlled system and preprocess the operating data to obtain a structured data stream; Digital twin model building module: Connected to the data acquisition and preprocessing module, it is used to filter structured data streams to obtain an optimized set of structured data streams, and to build a digital twin model using the optimized set of structured data streams; Closed-loop simulation system forming module: connected to the digital twin model building module, used to call pre-stored adaptive control algorithms and embed them into the digital twin model to form a closed-loop simulation system; Control execution module: It is connected to the closed-loop simulation system to form a module. It is used to run the simulation in the closed-loop simulation system with the current working condition as the initial condition, generate the optimized trajectory of control parameters, generate the control simulation strategy based on the optimized trajectory of control parameters, convert the control simulation strategy into control commands, and send them to the field controller to perform adjustment operations on the controlled system. The process of filtering structured data streams to obtain an optimized set of structured data streams, and then using this optimized set of structured data streams to construct a digital twin model, includes: Data classification and grouping unit: used to classify structured data streams according to data type and group each type of structured data stream; Correlation analysis unit: used to analyze the correlation between each group of structured data streams in each type of structured data stream, and obtain the correlation coefficient between each group of structured data streams in each type of structured data stream; The comprehensive correlation coefficient acquisition unit is used to synthesize the correlation coefficients between each group of structured data streams in each type of structured data stream to obtain the comprehensive correlation coefficient of each type of structured data stream. Importance Analysis Unit: Used to analyze the importance of each group of structured data streams in each type of structured data stream, and obtain the importance score of each group of structured data streams in each type of structured data stream; Comprehensive Importance Score Acquisition Unit: Used to synthesize the importance scores of each group of structured data streams in each type of structured data stream to obtain the comprehensive importance score of each type of structured data stream; The comprehensive index acquisition unit is used to normalize the comprehensive correlation coefficient and comprehensive importance score of each type of structured data stream, and combine the normalization results to obtain the comprehensive index of each type of structured data stream. Screening and Judgment Unit: Used to screen each group of structured data streams in each type of structured data stream by using the comprehensive correlation coefficient, comprehensive importance score, comprehensive index, correlation coefficient between each group of structured data streams in each type of structured data stream, and importance score of each group of structured data streams in each type of structured data stream, to obtain the optimized structured data streams in each type of structured data stream. Digital twin model building unit: used to aggregate each type of optimized structured data stream to obtain an optimized set of structured data streams, and to build a digital twin model using the optimized set of structured data streams.

2. The adaptive control simulation system based on a digital twin model according to claim 1, characterized in that, The analysis of the correlation between each group of structured data streams within each type of structured data stream yields the correlation coefficient between each group of structured data streams within each type of structured data stream, including: Numerical sequence acquisition subunit: used to acquire the numerical sequences of any two sets of structured data streams in each type of structured data stream within the same time interval, denoted as the first numerical sequence and the second numerical sequence, wherein the first numerical sequence contains T sampling points arranged in chronological order, and the second numerical sequence contains T sampling points at the corresponding time points; Numerical sequence calculation subunit: used to calculate the arithmetic mean of the first numerical sequence, which is calculated by dividing the sum of the values ​​of all sample points in the sequence by the total number of sample points T; and to calculate the arithmetic mean of the second numerical sequence, which is calculated by dividing the sum of the values ​​of all sample points in the sequence by the total number of sample points T. Covariance calculation subunit: used to calculate the covariance between the first numerical sequence and the second numerical sequence. The calculation method is as follows: for each time point t, calculate the product of the difference between the first value and the arithmetic mean of the first numerical sequence and the difference between the second value and the arithmetic mean of the second numerical sequence at that time point, and sum the products of all time points and divide by the total number of sampling points T. The standard deviation calculation subunit for the numerical sequence is used to calculate the standard deviation of the first numerical sequence. The calculation method is as follows: for each time point t, calculate the square of the difference between the first value at that point and the arithmetic mean of the first numerical sequence, sum the results, divide by T, and then take the square root. The standard deviation of the second numerical sequence is calculated as follows: for each time point t, calculate the square of the difference between the second value at that point and the arithmetic mean of the second numerical sequence, sum the results, divide by T, and then take the square root. Pearson correlation coefficient calculation subunit: This unit divides the covariance obtained from the covariance calculation subunit by the product of the two standard deviations obtained from the numerical sequence standard deviation calculation subunit to obtain the Pearson correlation coefficient between the two sets of structured data streams. The correlation coefficient calculation subunit is used to perform pairwise numerical sequence acquisition subunit to Pearson correlation coefficient calculation subunit on all data groups in the class to obtain the correlation coefficient between each group of structured data streams in each class of structured data streams.

3. The adaptive control simulation system based on a digital twin model according to claim 1, characterized in that, The analysis determines the importance of each group of structured data streams within each type of structured data stream, resulting in an importance score for each group of structured data streams within each type of structured data stream, including: Output response change acquisition subunit: used to apply a disturbance signal of preset amplitude to any set of structured data streams in each type of structured data stream, and acquire the output response change of the controlled system during the disturbance period; Control sensitivity value acquisition subunit: used to calculate the ratio of the change in the output response to the amplitude of the disturbance signal, and obtain the control sensitivity value of the structured data stream of the current group; Each control sensitivity value acquisition subunit: is used to execute the output response change acquisition subunit to the control sensitivity value acquisition subunit for each group of structured data streams to obtain the control sensitivity value of each group of structured data streams; Information entropy value acquisition sub-unit: used to acquire time series data of each group of structured data streams, divide the time series data of each group of structured data streams into N consecutive subsequences, calculate the information entropy of each subsequence, and then calculate the average of the information entropy of all subsequences to obtain the information entropy value of the current group of structured data streams; Information entropy score acquisition subunit: used to perform range normalization processing on the control sensitivity value and information entropy value to obtain the normalized control sensitivity score and the normalized information entropy score; Importance score acquisition subunit: used to take the arithmetic mean of the normalized control sensitivity score and information entropy score to obtain the importance score of the structured data stream in the current group. The information entropy score acquisition subunit is executed sequentially for each group of structured data streams to obtain the importance score of each group of structured data streams in each type of structured data stream.

4. The adaptive control simulation system based on a digital twin model according to claim 1, characterized in that, The optimized structured data streams are obtained by filtering each group of structured data streams within each type of structured data stream using the comprehensive correlation coefficient, comprehensive importance score, comprehensive index, correlation coefficient between each group of structured data streams within each type of structured data stream, and importance score of each group of structured data streams within each type of structured data stream. This includes: Threshold setting subunit: used to set the first threshold, second threshold, third threshold, fourth threshold and fifth threshold for the comprehensive index, comprehensive correlation coefficient, comprehensive importance score, correlation coefficient between each group of structured data streams in each type of structured data stream, and importance score of each group of structured data streams in each type of structured data stream, respectively. The comprehensive index determination subunit is used to determine the comprehensive correlation coefficient and comprehensive importance score if the comprehensive index of the current category of structured data stream is less than the first threshold; if the comprehensive index of the current category of structured data stream is greater than or equal to the first threshold, it will proceed to the data reassembly subunit. The comprehensive correlation coefficient and comprehensive importance score determination subunit is used to determine whether the current category's structured data stream has a comprehensive correlation coefficient less than the second threshold or a comprehensive importance score less than the third threshold. If the current category's structured data stream has a comprehensive correlation coefficient greater than or equal to the second threshold and a comprehensive importance score greater than or equal to the third threshold, it will proceed to the data reassembly subunit. The correlation coefficient and importance score determination subunit is used to remove structured data streams of the current group from the current category's structured data stream if the correlation coefficient between structured data streams of the current category and the current group is less than the fourth threshold or the importance score of structured data streams of the current group from the current category's structured data stream is less than the fifth threshold; if the correlation coefficient between structured data streams of the current category and the current group is greater than or equal to the fourth threshold and the importance score of structured data streams of the current group from the current category's structured data stream is greater than or equal to the fifth threshold, then the data reassembly subunit is entered. Data Reassembly Subunit: Used to reassemble the group of structured data streams that were not removed, to obtain optimized structured data streams for each type.

5. The adaptive control simulation system based on a digital twin model according to claim 4, characterized in that, The method sets a first threshold, a second threshold, a third threshold, a fourth threshold, and a fifth threshold for the comprehensive index, comprehensive correlation coefficient, comprehensive importance score, correlation coefficient between groups of structured data streams in each type of structured data stream, and importance score of each group of structured data streams in each type of structured data stream, respectively, including: The first threshold calculation interface is used to perform range normalization on the comprehensive correlation coefficient and comprehensive importance score of the structured data stream of the current category, respectively, to obtain the normalized comprehensive correlation coefficient and the normalized comprehensive importance score. The ratio of their product to their arithmetic mean is then used as the coupling strength factor. The 75th percentile of the coupling strength factor under normal operating conditions within the historical operating cycle is used as the coupling strength benchmark value. If the current coupling strength factor is greater than the coupling strength benchmark value, the minimum value between the normalized comprehensive correlation coefficient and the normalized comprehensive importance score is used as the first threshold; otherwise, the geometric mean of the two is used as the first threshold. The second threshold calculation interface is used to obtain the correlation coefficients between all groups in the current category and calculate the skewness and kurtosis of their probability distribution. The 90th percentile of the skewness of the same data stream in the historical running period is used as the skewness threshold, and the 90th percentile of the kurtosis is used as the kurtosis threshold. If the current absolute value of skewness is greater than the skewness threshold and the current kurtosis is greater than the kurtosis threshold, the maximum correlation coefficient is multiplied by 0.8 as the second threshold; otherwise, the median of the correlation coefficient is multiplied by 1.1 as the second threshold. The third threshold calculation interface is used to obtain the importance scores of all groups in the current category and calculate their coefficient of variation. The median of the coefficient of variation of the same type of structured data stream in the historical running period is used as the coefficient of variation threshold. If the current coefficient of variation is greater than the threshold, the lower quartile of the importance score is used as the third threshold. Otherwise, the arithmetic mean of the importance scores minus twice the standard deviation is used as the third threshold. The fourth threshold calculation interface is used to obtain the correlation coefficients between all groups in each type of structured data stream within the most recent N historical running cycles, construct the first sliding time window dataset, calculate the mean and standard deviation of the first sliding time window dataset, and determine the lower limit of the 95% confidence interval based on the t-distribution, and use this lower limit as the fourth threshold. The fifth threshold calculation interface is used to obtain the importance scores of all groups in each type of structured data stream within the most recent N historical running cycles, construct the second sliding time window dataset, calculate the mean and standard deviation of the second sliding time window dataset, and determine the lower limit of the 95% confidence interval based on the t-distribution, and use this lower limit as the fifth threshold.

Images