Wide-working-condition steam turbine system operation data correction method based on digital twinning

Abstract

The invention discloses a wide-working-condition steam turbine system operation data correction method based on digital twinning, and relates to the technical field of steam turbine equipment digitalization, and the method comprises the steps: collecting operation data of a wide-working-condition steam turbine system in real time; constructing a digital twinborn model of the steam turbine system; performing prior covariance estimation on the operation data according to the statistical distribution condition and the design parameters; generating a Sigma point by adopting an unscented Kalman filtering algorithm, iteratively updating a state vector and a covariance in combination with a state transition equation and an observation equation, and outputting an updated known variable; performing smoothing processing on the updated known variable in a plurality of time point regions by adopting an RLOESS algorithm to obtain corrected operation data; according to the method, the real dynamic characteristics of the wide-working-condition steam turbine are approached by building the digital twinborn model, the calculation complexity of dynamic data fusion of a steam turbine system can be remarkably reduced, and the problems that the convergence speed is low and calculation is complex due to strong nonlinearity of a traditional method are solved.

Description

Technical Field

[0001] This invention relates to the field of digital technology for steam turbine equipment, and specifically to a method for correcting operating data of a steam turbine system under wide operating conditions based on digital twins. Background Technology

[0002] Against the backdrop of achieving the "dual carbon" target, large thermal power generating units are participating more frequently in deep grid peak shaving, making the wide-condition steam turbine, as its core equipment, particularly important. Compared to traditional application scenarios, the operating conditions of wide-condition steam turbines change more frequently, thus placing higher demands on their operation and maintenance monitoring systems. In order to build an intelligent operation and maintenance framework for wide-condition steam turbine systems with complex physical constraints, high nonlinearity, and complex dynamic adjustment characteristics, more effective data processing and correction methods are needed.

[0003] In existing methods, the dynamic data coordination method fuses the system state transition model with actual measurement data using a Kalman filter, and corrects the measurement data with errors through physical equilibrium equations and other constraint relationships, thereby tracking the trend of system state changes more accurately.

[0004] However, existing methods typically rely on simplified models based on empirical formulas, which make it difficult to balance model accuracy and computational efficiency. They are not robust enough to measurement noise and long-term data gaps, and they cannot make full use of temporal redundancy information, only using data from the previous moment to correct the current result, resulting in low reliability of the data correction results. Summary of the Invention

[0005] To address the shortcomings of existing technologies that cannot fully utilize time redundancy information and rely solely on data from the previous moment to correct the current result, resulting in low reliability of the corrected data, this invention proposes a wide-condition steam turbine system operation data correction method based on digital twins, thereby solving the problems existing in the prior art.

[0006] A method for correcting operating data of a steam turbine system under wide operating conditions based on digital twins includes the following steps: Real-time acquisition of operating data from a wide range of steam turbine systems; Input and output boundary condition variables with a coupling degree higher than a set threshold with the dynamic characteristics of the steam turbine system under wide operating conditions are selected from the operating data to construct a digital twin model of the steam turbine system. Based on the statistical distribution characteristics and design parameters of the turbine system operating data, prior covariance estimation is performed on the operating data to obtain the initial covariance matrix of the operating data; The digital twin model of the steam turbine system is converted into state transition equations to generate state vectors. Using operating data as known variables, the observation equations of the steam turbine system are obtained by establishing functional relationships between the known variables and the state vectors. Based on the observation equations and the initial covariance matrix, the covariance matrix of the state variables is estimated. Based on the initial covariance matrix of the operating data, an unscented Kalman filter algorithm is used to generate Sigma points, which are then combined with the state transition equations and the observation equations to iteratively update the state vectors and their covariance matrices until the updated known variables are output. The locally weighted regression algorithm is used to smooth the updated known variables over multiple time periods to obtain the corrected running data.

[0007] Furthermore, the step of selecting input and output boundary condition variables from the operating data whose coupling degree with the dynamic characteristics of the steam turbine system under wide operating conditions is higher than a set threshold, and constructing a digital twin model of the steam turbine system, specifically includes the following steps: Input and output boundary condition variables in the operating data whose coupling degree with the dynamic characteristics of the wide-condition steam turbine system is higher than a set threshold are selected to generate a wide-condition steam turbine boundary condition dataset. Based on the dynamic characteristics of the steam turbine system, it is divided into multiple steam turbine modules; based on the wide-condition steam turbine boundary condition dataset, a digital twin model is constructed for each steam turbine module that can capture the dynamic relationship between its input and output boundary condition variables; By integrating the digital twin models of each turbine module, a digital twin model of the entire turbine system is constructed.

[0008] Furthermore, the digital twin model is constructed using the Kriging response surface model, support vector regression (SVR), recurrent neural network (RNN), or physically driven neural network (PINN).

[0009] Furthermore, it also includes defining an implicit function after constructing a digital twin model of the turbine system. : At a given moment In the case of known parameters Solve for all unknown parameters in the steam turbine system .

[0010] Furthermore, based on the statistical distribution characteristics and design parameters of the turbine system operating data, prior covariance estimation is performed on the operating data to obtain the initial covariance matrix of the operating data; specifically, this includes the following steps: The first in the calculation running data Normalized parameters The variance estimate; where, when the parameter When obtained through instrument measurement, its variance estimate is expressed as: in, For parameters The variance estimate, for In the The original measurements at each time point for all The mean of all measured values; Parameters are estimated and determined using turbine design data. Based on engineering experience, Setting uncertain initial values Based on the 95% confidence interval width, Convert to variance, and get The variance estimate: By summing the variance estimates of each parameter in the turbine system operating data, the initial covariance matrix of the operating dataset is obtained: ; in, n For the number of times measured.

[0011] Furthermore, the process of converting the digital twin model of the steam turbine system into state transition equations and generating state vectors is expressed as follows: ; in For the first The state vector at each time point For control vectors, This is the state transition function. For process noise that follows a zero-mean multivariate normal distribution N, its covariance matrix is ​​defined as follows. ,but: .

[0012] Furthermore, by using the operating data as known variables and establishing a functional relationship between the known variables and the state vector, the observation equations of the turbine system are obtained, which are expressed as follows: ; in, Given a vector of variables, For the observation function, For observation noise that follows a zero-mean multivariate normal distribution N, its covariance matrix is ​​defined as follows. ,but .

[0013] Furthermore, the initial covariance matrix based on the running data is used to generate Sigma points using an unscented Kalman filter algorithm, and these points are combined with the state transition equation and the observation equation to iteratively update the state vector and its covariance matrix until the updated known variables are output. Specifically, this includes the following steps: Estimate the covariance matrix of the state variables based on the observation equation: ;in For known variables Regarding state variables Jacobian matrix; Define the initial value of the state vector as follows: Its dimensions are Perform an unscented transformation on it to generate Sigma points: ;in To control the parameters of the Sigma point distribution, and All are given adjustment parameters; Represents a matrix After finding the square root matrix, take its nth... Column vector; right The propagation results of each Sigma point are estimated posteriorly, and the propagation results of each Sigma point are summed according to the set weights to obtain the final result. Calculate the weight vector of the mean of each Sigma point : ; Then the weight vector of the covariance at each Sigma point for: ;in These are parameters used to introduce prior distribution information; Propagation is performed on each Sigma point to obtain prior estimates of the mean and covariance of each Sigma point. The propagation result of the mean, combined with the state transition equation, is expressed as follows: ;in, Indicates the first At the 1st time point Prior estimates of Sigma points Indicates the first The posterior estimate of the Sigma point at the previous time step, when The time is taken as the initial value ; pass The weights of the Sigma points are combined to form the prior estimate of the state vector at the current time point: Then the prior estimate of the covariance at the current time point is calculated as follows: ; Based on the unscented Kalman filter algorithm, estimate the observation data generated at each Sigma point: The covariance matrix of the observed data is calculated as follows: The cross-covariance between the state vector and the observation vector is then expressed as: ; Based on the principle of the Kalman filter algorithm, the Kalman gain under the current state variable and the observed data is calculated as follows: ; By performing posterior estimation on the state vector, the updated state variables are obtained as follows: The updated covariance after posterior time is then expressed as: ; After updating the state vector, update the observations of the known variables to obtain the updated known variables: .

[0014] Furthermore, the local weighted regression algorithm is used to smooth the updated known variables across multiple time points to obtain corrected operational data; specifically, this includes the following steps: For the updated known variables Define any one of its components These are the initial observation parameters; Define bandwidth parameters based on the magnitude of abnormal errors in the operational data. This parameter represents the proportion of data used in the example; For the initial observation parameters, define the time axis closest to them. Each sample point is used as the corresponding time window, A local weighted regression is performed on each sample point as a data point; the data within the time window are recorded. The points are The regression weights are defined as follows: ;in, within the time window The distance to the farthest point is then used within the time window. A polynomial fit of order 1 to all data points, including the initial observation parameters, yields: ;in, for The fitted value, for The coefficients of a polynomial of order 1; Solve for the optimal values ​​of the coefficients of polynomials of each order: ;in This is the vector of polynomial coefficients; Calculate the regression residuals for each data point : ; definition Stable weighting at points for: ;in, For all regression residuals the median; pass The robust weights at the point are updated to their regression weight parameters, resulting in the updated Euclidean algorithm. The regression weights of the points are: ; The updated regression weights are substituted into the process of solving for the optimal values ​​of the polynomial coefficients of each order to determine the optimized polynomial coefficient vector. ; After multiple iterations, the initial weights in two adjacent steps With updated weights If the relative change is less than a given threshold, the iterative process converges, and the current updated weights are output. Update the current weight Substitute During the process of fitting a polynomial of order 1, the following is obtained: The corresponding predicted data point This will be used as the corrected running data.

[0015] Furthermore, after acquiring the real-time operating data of the wide-condition steam turbine system, the system performs preprocessing on the operating data. The preprocessing process includes the following steps: Based on the thermal design data, equipment design data, and physical meaning of the parameters of the steam turbine system, each measured operating parameter in the steam turbine system is analyzed. Determine its lower limit value Upper limit With design value ; For any given moment Operational data measurement results To determine the potential problem types, including: like If measurement results are missing, the type of missing data should be identified. For short-term missing data, nearest neighbor interpolation should be used to fill the missing values. For long-term missing data, the design values ​​should be used directly. Replace the current missing measurement value; and mark the current parameter as a missing parameter; like If no measurement results are missing, then it is determined whether the measurement results are abnormal; among them, when the measurement results are outside the limit range... If the measurement results are abnormal, the design value will be used directly. Replace the original measurement result and mark the current parameter as an outlier; After cleaning up missing and outlier values ​​in the measurement results of the operational data, the Min-Max method is used to normalize the processed operational data to obtain the preprocessed operational data.

[0016] This invention provides a method for correcting operating data of a steam turbine system under wide operating conditions based on digital twins, which has the following beneficial effects: This invention approximates the true dynamic characteristics of a steam turbine under a wide operating condition by constructing a high-precision digital twin model. This significantly reduces the computational complexity of dynamic data fusion for the steam turbine system, solving the problems of slow convergence and computational complexity caused by strong nonlinearity in traditional methods. It combines statistical distribution and design parameters for covariance estimation, improving tolerance to noise and missing data. Furthermore, by incorporating an unscented Kalman filter algorithm, it effectively fuses the digital twin model with real-time measurement data, accurately correcting abnormal and missing data generated by sensors while suppressing measurement errors. Simultaneously, it uses a local weighted regression algorithm to robustly smooth the data across multiple time windows, enhancing the temporal stability of the fused data and improving the reliability of the data correction results. This provides reliable data support for an intelligent operation and maintenance platform for steam turbines under a wide operating condition. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for correcting operating data of a steam turbine system under wide operating conditions based on digital twins, as described in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] This invention proposes a method for correcting operating data of a steam turbine system under wide operating conditions based on digital twins, such as... Figure 1 As shown, the method specifically includes the following steps: S1. Obtain real-time operating data of steam turbines under wide operating conditions.

[0020] During operation, steam turbines and their auxiliary systems generate various forms of operational data. Measuring instruments within the steam turbine system include those for pressure, temperature, flow rate, and power, and are organized using monitoring / control systems such as SCADA, DCS, and PLC to continuously collect real-time data streams during the system's dynamic operation. Let the set of time points corresponding to all valid data be denoted as... At every point in time It can obtain operating data under the current operating conditions from the measurement system and arrange all operating parameters of the system into a vector:

[0021] in, Let be the number of measurable parameters in the system; and for any... , Indicates the first Each running parameter is Measurement results at any given time. The data may contain missing or outliers.

[0022] S2, Data Processing.

[0023] The runtime data may contain missing or outlier values ​​at a certain point in time. The running parameter vector This involves identifying and processing missing and outlier values. First, based on reference information such as system thermal design data, equipment design data, and the physical meaning of parameters, each measurable operating parameter in the system is analyzed. Determine its lower limit value Upper limit With design value Subsequently, for any given moment... Operating parameter measurement results To determine the potential types of problems.

[0024] like If measurement results are missing, the type of missing data should be identified. If the missing data is for a short period, nearest neighbor interpolation (or linear interpolation) should be used to fill the missing values. If the missing data is for a long period, the design values ​​should be used directly. Replace the current missing measure. Mark the current parameter as a missing parameter in the dataset.

[0025] Assuming no measurement results are missing, further determine if there are any anomalies in the measured values. If the measurement results are outside the limit range, that is... If the measurement results are significantly abnormal, the design value should be used directly. Replace the original measurement value and mark the current parameter as an outlier in the dataset.

[0026] After cleaning up missing and outlier values, the real-time data is normalized to eliminate the influence of different physical dimensions and data ranges on subsequent data fusion algorithms. Specifically, for any parameter measurement result at any given time after cleaning... Normalization was performed using the Min-Max method:

[0027] Among them The normalized operating parameter measurement results, assuming normal data cleaning is completed, will always have .

[0028] S3. Construct a wide-condition steam turbine boundary condition dataset.

[0029] Identify key boundary condition parameters that significantly influence the dynamic characteristics of a wide-range steam turbine. These parameters are then broken down according to their respective modules—including turbine stages, peripheral heat exchangers, valves, and pressure-loss components such as long-flow piping—and defined as input and output boundary condition variables for the wide-range steam turbine. This ensures that they fully describe the interaction between the turbine and other parts of its thermodynamic system. For example, for a single, independently measurable stage in the turbine (typically defined by the inlet, outlet, and extraction points), define its inlet mass flow rate, pressure, temperature, and shaft speed as input boundary variables, and its outlet pressure and temperature as output boundary condition variables. A clear one-to-one correspondence should exist between input and output variables, which can be determined through turbine design data, numerical simulations (one-dimensional heat exchanger design calculations, computational fluid dynamics analysis), etc. If the wide-range steam turbine can be broken down according to the above principles... There are several modules with input and output boundary condition variables, where the boundary condition dataset of one module at any given time is:

[0030] Where, for any , Indicates the first A subset of boundary conditions for each module, containing The values ​​of all boundary condition variables of the device at the given time. This represents the set of virtual time points, which corresponds to the set of real time points in the system's operational data. Unlike other methods, this one generates data directly based on the needs of subsequent digital twin model training. Correspondingly, for the first... Each module has its boundary condition dataset defined over the entire virtual time point set as follows:

[0031] All variables in the boundary condition dataset should be normalized according to the Min-Max method defined in the formula to meet the requirements of building a digital twin model in subsequent steps.

[0032] S4. Construct digital twin models of each module of the steam turbine system.

[0033] Regarding the steps mentioned above, For each module, select and construct a digital twin model that can accurately and efficiently capture the dynamic relationship between its input-output boundary conditions. Possible approaches to constructing digital twin models include Kriging response surface methodology, Support Vector Regression (SVR), Recurrent Neural Networks (RNN), and Physically Driven Neural Networks (PINN), etc., and the optimal approach should be selected based on the number of input / output parameters, dataset size, nonlinear characteristics, parameter sensitivity, and other characteristics of each module. For example, for a relatively simple module like the pressure loss component, a Kriging response surface model or an SVR model can be used, while for the turbine stage group, a neural network model should be used to construct its digital twin model. Specifically, for the first... Each module, with its boundary condition dataset arranged in an appropriate proportion. The dataset is divided into training, validation, and test sets, and an appropriate training method is used to construct a digital twin model. After training, the accuracy of the digital twin model should be validated on the designated test set to ensure that its mean squared error remains at a low level (on a normalized dataset, the mean squared error should decrease to a certain level). After training, the mapping relationship between the input boundary conditions and the output boundary conditions of the module can be obtained through the digital twin model.

[0034] S5. Construct a digital twin model of a steam turbine system under a wide range of operating conditions.

[0035] For the wide-condition steam turbine under study, the digital twin models of each module trained in S4 are connected to construct a high-precision digital twin model of the entire steam turbine system. The various parameters in the system are divided into two categories: sensor parameters measured by arranging measuring points and system operating parameters defined according to design values, which are defined as an operating parameter vector by the formula. Each component, after being normalized according to the formula, is defined as a known parameter vector:

[0036] in, n The number of measurements is denoted by 'measurement number'; other parameters whose values ​​cannot be determined are called unknown parameters, and are arranged as a vector. The function of a digital twin computational model for a steam turbine system is to perform calculations at a given time. Based on known parameters Solving for unknown parameters This process can be expressed as an implicit function. :

[0037] To solve the digital twin model described by implicit functional equations, it is necessary to comprehensively consider the digital twin models of each module in the wide-condition steam turbine and other physical constraints. The above multivariate equations hold true when both known and unknown parameters are accurate values, satisfying the implicit functional equations. Otherwise, if the residuals are not zero, it indicates errors in the measured values ​​of known parameters or the estimated values ​​of unknown parameters, requiring correction. For the wide-condition steam turbine digital twin model constructed according to the above steps, it should be ensured that the known parameters... Given a given condition, all unknown parameters within the turbine system can be solved using the formula. .

[0038] S6. Make prior estimates of the covariance of the running data.

[0039] Based on the statistical distribution and design data, prior covariance estimation is performed on the operational data. For the first... Normalized parameters If it is obtained through instrument measurement, its variance is estimated unbiasedly using the following formula:

[0040] in, For parameters The variance estimate, For its in the The original measurements at each time point For all of this parameter The mean of all measured values. If it is estimated using design data or other methods, its variance cannot be estimated using the formula because its initial value usually remains constant at each time point. In this case, a higher initial value with uncertainty can be set based on engineering experience. (e.g., 20% or 30%), convert it to variance using a 95% confidence interval width:

[0041] Based on this, a priori estimation of the system covariance can be performed, and the variance estimates of each component can be arranged to obtain the initial covariance matrix of the running dataset: .

[0042] S7. Use the unscented Kalman filter algorithm for posterior state estimation.

[0043] To integrate the system's digital twin model with operational data, and especially to identify and suppress unknown errors in the latter, an unscented Kalman filter algorithm is used to estimate the system's posterior state. Specifically, firstly, several Sigma points need to be generated from the original measurement data through an unscented transformation. Therefore, the system's digital twin model, as described in the equation, is rewritten as the state transition equation shown below:

[0044] in For the first The state vector at each time point For control vectors, This is the state transition function. The process noise follows a zero-mean multivariate normal distribution, and its covariance matrix is ​​assumed to be... , that is: To determine the system state vector Establish a system based on known variables The functional relationship between the system and the state vector is called the system's observation equation: in, Given a vector of variables, For the observation function, For observation noise that follows a zero-mean multivariate normal distribution, and assume its covariance matrix is... , that is In the above state transition equations and observation equations and All are nonlinear functions. Furthermore, the covariance matrix of the state variables can be estimated based on the observation equations:

[0045] in The known variables obtained by differentiating the expression Regarding state variables The Jacobian matrix. Let the initial value of the state vector be... Its dimensions are Perform an unscented transformation on it to generate Sigma points:

[0046] in To control the parameters of the Sigma point distribution, and All of these are manually assigned adjustment parameters; Represents a matrix Find the square root matrix (using methods such as Cholesky decomposition) and then take its nth root. Column vectors. In subsequent calculations, these need to be processed separately. A posterior estimate is performed on each Sigma point, and then the propagation results of each Sigma point are summed according to certain weights to form the final result. Regarding the calculation of the weights, the weight vector corresponding to the mean of each Sigma point is... Determine by the following formula:

[0047] The weight vector corresponding to the covariance of each Sigma point Determine by the following formula: in For parameters used to introduce prior distribution information, it is usually assumed that the prior distribution is a normal distribution, and the parameters are taken as follows: Based on this, propagation is performed on each Sigma point to obtain prior estimates of their mean and covariance. The propagation result of the mean is calculated according to the formula, and a wide-condition steam turbine digital twin model is applied in the process:

[0048] In this formula, Indicates the first At the 1st time point Prior estimates of Sigma points Indicates the first The posterior estimate of the Sigma point at a given time point, when The time is taken as the initial value Otherwise, it is determined iteratively through subsequent calculation steps. (From all) The prior estimate of the state vector at the current time point is synthesized from the Sigma points according to the aforementioned weights:

[0049] Meanwhile, the prior estimate of the covariance at the current time point is calculated using the following formula: Based on the unscented Kalman filter algorithm, the observed data generated at each Sigma point are first estimated: The covariance matrix of the observed data is calculated as follows: On the other hand, the cross-covariance between the state vector and the observation vector can be calculated as follows: Therefore, based on the principle of the Kalman filter algorithm, the Kalman gain under the current state variable and the observed data is calculated as follows: Therefore, a posteriori estimation can be performed on the state vector to obtain the updated state variables: Simultaneously, a posterior update of the covariance is performed, yielding: After updating the state vector, the observations of known variables can be updated based on this, thus completing the dynamic data coordination analysis process. in, This represents the updated known variables.

[0050] S8. Apply the local weighted regression algorithm to smooth the results.

[0051] The data processed by the unscented Kalman filter algorithm can effectively fuse the features of the model and the original measurement data. However, it may perform poorly when there are large abnormal fluctuations in the measurement data. Therefore, the robust locally estimated scatterplot smoothing (RLOESS) algorithm is further applied to smooth the fused data over multiple time periods to improve the stability of the fusion result in long-term samples.

[0052] Specifically, for known variables corrected using the unscented Kalman filter algorithm Consider any one of its components These are called the initial observation parameters. The bandwidth parameter is defined based on the magnitude of the outlier errors in the data. This parameter represents the proportion of data used for fitting. For the initial observation parameters, the data closest to them on the time axis is defined. Each sample point is used as a corresponding time window, and local weighted regression is performed using these points as data points. Let the time window contain... The points are The regression weight parameters for each are defined as follows:

[0053] in, within the time window The distance to the farthest point. Based on this, within the time window, [the following is used]. The polynomial fitting of order one includes all data points, including the initial observation parameters:

[0054] in, for The fitted value, for The coefficients of polynomials of order X. To find the optimal values ​​of the coefficients of polynomials of order X, a weighted least squares optimization problem is constructed as follows:

[0055] in Let be the polynomial coefficient vector. Solving this optimization problem yields the initial weighted regression polynomial, but the result obtained is susceptible to outlier values. Therefore, it is necessary to further calculate the regression residuals for each data point. :

[0056] Therefore, by definition Stable weighting at points for: in, For all regression residuals The median. Using the robust weights calculated above, the initial weights are updated according to the formula to obtain the corresponding... The update weight of the point is:

[0057] By substituting the updated weights into the equation to replace the original weight coefficients, the weighted least squares optimization problem is redefined and solved to determine a better polynomial coefficient vector. And from this, a better prediction data point is obtained according to the formula. Repeat the above iterative process, if the initial weights in adjacent steps... With updated weights If the relative change is less than a given threshold, the iterative process is considered to have converged, the calculation stops, and the current updated weights are taken. This is the final weight to be used. Substituting it into the formula, we obtain... The corresponding predicted data point The final fusion result of this parameter is used as the result.

[0058] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for correcting operating data of a steam turbine system under wide operating conditions based on digital twins, characterized in that, Includes the following steps: Real-time acquisition of operating data from a wide range of steam turbine systems; Input and output boundary condition variables with a coupling degree higher than a set threshold with the dynamic characteristics of the steam turbine system under wide operating conditions are selected from the operating data to construct a digital twin model of the steam turbine system. Based on the statistical distribution characteristics and design parameters of the turbine system operating data, prior covariance estimation is performed on the operating data to obtain the initial covariance matrix of the operating data; The digital twin model of the steam turbine system is converted into state transition equations to generate state vectors; the operating data is used as known variables, and the observation equations of the steam turbine system are obtained by establishing the functional relationship between the known variables and the state vectors. The covariance matrix of the state variables is estimated based on the observation equation and the initial covariance matrix; Based on the initial covariance matrix of the running data, the unscented Kalman filter algorithm is used to generate Sigma points, which are then combined with the state transition equation and the observation equation to iteratively update the state vector and its covariance matrix until the updated known variables are output. The locally weighted regression algorithm is used to smooth the updated known variables over multiple time points to obtain the corrected running data.

2. The method for correcting operating data of a steam turbine system under wide operating conditions based on digital twins according to claim 1, characterized in that, The process of selecting input and output boundary condition variables from the operating data whose coupling degree with the dynamic characteristics of the steam turbine system under wide operating conditions exceeds a set threshold in order to construct a digital twin model of the steam turbine system includes the following steps: Input and output boundary condition variables in the operating data whose coupling degree with the dynamic characteristics of the wide-condition steam turbine system is higher than a set threshold are selected to generate a wide-condition steam turbine boundary condition dataset; Based on the dynamic characteristics of the steam turbine system, it is divided into multiple steam turbine modules; based on the wide-condition steam turbine boundary condition dataset, a digital twin model is constructed for each steam turbine module that can capture the dynamic relationship between its input and output boundary condition variables; By integrating the digital twin models of each turbine module, a digital twin model of the entire turbine system is constructed.

3. The method for correcting operating data of a steam turbine system under wide operating conditions based on digital twins according to claim 1, characterized in that, The digital twin model is constructed using the Kriging response surface model, support vector regression (SVR), recurrent neural network (RNN), or physical driven neural network (PINN).

4. The method for correcting operating data of a steam turbine system under wide operating conditions based on digital twins according to claim 1, characterized in that, This also includes defining an implicit function after constructing the digital twin model of the steam turbine system. : At a given moment In the case of known parameters Solve for all unknown parameters in the steam turbine system .

5. The method for correcting operating data of a wide-condition steam turbine system based on digital twins according to claim 1, characterized in that, Based on the statistical distribution characteristics and design parameters of the turbine system operating data, the prior covariance of the operating data is estimated to obtain the initial covariance matrix of the operating data. Specifically, the following steps are included: The first in the calculation running data Normalized parameters The variance estimate; where, when the parameter When obtained through instrument measurement, its variance estimate is expressed as: ； in, For parameters The variance estimate, for In the The original measurements at each time point for all The mean of all measured values; Parameters are estimated and determined using turbine design data. Based on engineering experience, Setting uncertain initial values Based on the 95% confidence interval width, Convert to variance, and get The variance estimate: ； By summing the variance estimates of each parameter in the turbine system operating data, the initial covariance matrix of the operating dataset is obtained: ； in, n For the number of times measured.

6. The method for correcting operating data of a steam turbine system under wide operating conditions based on digital twins according to claim 1, characterized in that, The process of converting the digital twin model of the steam turbine system into state transition equations and generating state vectors is expressed as follows: ； in For the first The state vector at each time point For control vectors, This is the state transition function. For process noise that follows a zero-mean multivariate normal distribution N, its covariance matrix is ​​defined as follows. ,but: .

7. The method for correcting operating data of a steam turbine system under wide operating conditions based on digital twins according to claim 6, characterized in that, The process involves using the operating data as known variables and establishing a functional relationship between the known variables and the state vector to obtain the observation equations of the turbine system. These observation equations are expressed as follows: ； in, Given a vector of variables, For the observation function, For observation noise that follows a zero-mean multivariate normal distribution N, its covariance matrix is ​​defined as follows. ,but .

8. The method for correcting operating data of a steam turbine system under wide operating conditions based on digital twins according to claim 7, characterized in that, The initial covariance matrix based on the running data is used to generate Sigma points using an unscented Kalman filter algorithm. These points are then combined with the state transition equation and the observation equation to iteratively update the state vector and its covariance matrix until the updated known variables are output. Specifically, this includes the following steps: Estimate the covariance matrix of the state variables based on the observation equation: ;in For known variables Regarding state variables Jacobian matrix; Define the initial value of the state vector as follows: Its dimensions are Perform an unscented transformation on it to generate Sigma points: ;in To control the parameters of the Sigma point distribution, and All are given adjustment parameters; Represents a matrix After finding the square root matrix, take its nth... Column vector; right The propagation results of each Sigma point are estimated posteriorly, and the propagation results of each Sigma point are summed according to the set weights to obtain the final result. Calculate the weight vector of the mean of each Sigma point : ; Then the weight vector of the covariance at each Sigma point for: ;in These are parameters used to introduce prior distribution information; Propagation is performed on each Sigma point to obtain prior estimates of the mean and covariance of each Sigma point. The propagation result of the mean, combined with the state transition equation, is expressed as follows: ;in, Indicates the first At the 1st time point Prior estimates of Sigma points Indicates the first The posterior estimate of the Sigma point at the previous time step, when The time is taken as the initial value ; pass The weights of the Sigma points are combined to form the prior estimate of the state vector at the current time point: Then the prior estimate of the covariance at the current time point is calculated as follows: ; Based on the unscented Kalman filter algorithm, estimate the observation data generated at each Sigma point: The covariance matrix of the observed data is calculated as follows: The cross-covariance between the state vector and the observation vector is then expressed as: ; Based on the principle of the Kalman filter algorithm, the Kalman gain under the current state variable and the observed data is calculated as follows: ; By performing posterior estimation on the state vector, the updated state variables are obtained as follows: The updated covariance after posterior time is then expressed as: ; After updating the state vector, update the observations of the known variables to obtain the updated known variables: .

9. The method for correcting operating data of a wide-condition steam turbine system based on digital twins according to claim 8, characterized in that, The process involves using a local weighted regression algorithm to smooth the updated known variables across multiple time points to obtain corrected operational data; specifically, it includes the following steps: For the updated known variables Define any one of its components These are the initial observation parameters; Define bandwidth parameters based on the magnitude of abnormal errors in the operational data. This parameter represents the proportion of data used in the example; For the initial observation parameters, define the time axis closest to them. Each sample point is used as the corresponding time window, A local weighted regression is performed on each sample point as a data point; the data within the time window are recorded. The points are The regression weights are defined as follows: ;in, within the time window The distance to the farthest point is then used within the time window. A polynomial fit of order 1 to all data points, including the initial observation parameters, yields: ;in, for The fitted value, for The coefficients of a polynomial of order 1; Solve for the optimal values ​​of the coefficients of polynomials of each order: ;in This is the vector of polynomial coefficients; Calculate the regression residuals for each data point : ; definition Stable weighting at points for: ;in, For all regression residuals the median; pass The robust weights at the point are updated to their regression weight parameters, resulting in the updated Euclidean algorithm. The regression weights of the points are: ; The updated regression weights are substituted into the process of solving for the optimal values ​​of the polynomial coefficients of each order to determine the optimized polynomial coefficient vector. ; After multiple iterations, the initial weights in two adjacent steps With updated weights If the relative change is less than a given threshold, the iterative process converges, and the current updated weights are output. Update the current weight Substitute During the process of fitting a polynomial of order 1, the following is obtained: The corresponding predicted data point This will be used as the corrected running data.

10. The method for correcting operating data of a wide-condition steam turbine system based on digital twins according to claim 1, characterized in that, It also includes preprocessing the operating data of the real-time wide-condition steam turbine system after acquisition. The preprocessing process includes the following steps: Based on the thermal design data, equipment design data, and physical meaning of the parameters of the steam turbine system, each measured operating parameter in the steam turbine system is analyzed. Determine its lower limit value Upper limit With design value ; For any given moment Operational data measurement results To determine the potential problem types, including: like If measurement results are missing, the type of missing data should be identified. For short-term missing data, nearest neighbor interpolation should be used to fill the missing values. For long-term missing data, the design values ​​should be used directly. Replace the current missing measurement value; and mark the current parameter as a missing parameter; like If no measurement results are missing, then it is determined whether the measurement results are abnormal; among them, when the measurement results are outside the limit range... If the measurement results are abnormal, the design value will be used directly. Replace the original measurement result and mark the current parameter as an outlier; After cleaning up missing and outlier values ​​in the measurement results of the operational data, the Min-Max method is used to normalize the processed operational data to obtain the preprocessed operational data.

Images