A full-factor real-time mapping digital twin simulation method and system for a hydropower station

Abstract

The application discloses a kind of hydropower station full-element real-time mapping digital twin simulation method and system, it is related to digital twin technical field, including the construction including hydraulic, mechanical, electrical and control state variable full-element unified state vector, and based on the current operating mode of identification of hydropower station operation data real-time acquisition;According to the current operating mode to the preset multi-domain reference coupling matrix is weighted and fused, generate the time-varying dynamic coupling operator for driving multi-domain coupling reduced mechanism model execution real-time deduction operation, obtain mechanism prediction output value;By constructing full-element unified state vector and generating time-varying dynamic coupling operator based on operating condition probability weighted fusion, the smooth continuous deduction of multi-domain coupling reduced mechanism model under complex mixed operating condition is realized, effectively eliminates the numerical oscillation caused by traditional hard switching, combined with residual correction network embedded physical conservation constraint and dynamic confidence adaptive fusion mechanism based on sensor real-time credibility.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, and in particular to a method and system for real-time mapping of all elements of a hydropower station using digital twin simulation. Background Technology

[0002] With the advancement of the energy internet and the construction of smart hydropower plants, digital twin technology for hydropower stations has become a core means to achieve unit condition monitoring, fault early warning, and optimized operation. Traditional modeling and simulation methods are mostly based on a single physical mechanism or purely data-driven approaches, which are difficult to cope with the complex and ever-changing operating environment of hydropower stations. Although mechanistic models have strong physical interpretability, they often suffer from parameter mismatch and decreased simulation accuracy when faced with equipment aging, frequent changes in operating conditions, and strong nonlinear coupling. On the other hand, purely data-driven models rely heavily on massive amounts of high-quality samples and are prone to outputting false results that violate physical conservation laws, lacking generalization robustness under extreme operating conditions.

[0003] Furthermore, existing digital twin systems often employ discrete operating condition switching logic when dealing with multi-domain coupling of hydraulics, mechanics, electrical systems, and control systems, which can easily lead to numerical oscillations at the boundaries. Simultaneously, traditional methods struggle to effectively distinguish between sensor noise and the actual system dynamics, and once model parameters are calibrated, they are difficult to update online adaptively as the equipment's performance evolves throughout its lifecycle. This results in a gradual disconnect between the virtual twin and the physical entity, failing to meet the requirements of high-fidelity real-time simulation. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a real-time mapping digital twin simulation method for all elements of a hydropower station. This solves the problem that existing digital twin systems, when dealing with multi-domain coupling of hydraulics, mechanics, electrical systems, and control, often employ discrete operating condition switching logic, which easily leads to numerical oscillations at boundaries. Furthermore, traditional methods struggle to effectively distinguish between sensor noise and the actual system dynamics, and once model parameters are calibrated, they are difficult to update online adaptively with the performance evolution throughout the equipment's lifecycle. This results in a gradual disconnect between the virtual twin and the physical entity, failing to meet the requirements of high-fidelity real-time simulation.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a real-time mapping digital twin simulation method for all elements of a hydropower station, comprising: Construct a unified state vector that includes hydraulic, mechanical, electrical, and control state variables, and identify the current operating mode based on real-time collected hydropower station operation data; Based on the current operating conditions, the pre-set multi-domain benchmark coupling matrix is ​​weighted and fused to generate a time-varying dynamic coupling operator for driving the multi-domain coupled reduced-order mechanism model to perform real-time inference operations, and the mechanism prediction output value is obtained. The difference between the measured data after spatiotemporal alignment and the mechanism prediction output is calculated to obtain the original residual sequence. The unified state vector of all elements, the real-time control input and the original residual sequence are then input into the residual correction network embedded with physical conservation constraints to obtain the simulation correction. Based on the sensor confidence evaluation results, a dynamic confidence factor is determined, and the dynamic confidence factor is used to weight and fuse the mechanism prediction output value and the simulation correction value to obtain the real-time output value of the high-fidelity twin. For each device performance parameter to be identified, the real-time excitation signal and the system response signal are selectively weighted in the frequency domain according to the feature identification frequency band, and the input-output cross-spectral coherence function is calculated. The product of the cross-spectral coherence function and the weighted spectral energy is integrally weighted exponentially within a preset sliding time window to obtain the effective energy density for parameter identification. Using the effective energy density identified by parameters as the modulation factor, a regularized recursive estimation algorithm is used to continuously and adaptively invert the performance parameters of the device to be identified to obtain time-varying parameter estimates. The time-varying parameter estimates are then injected into the physical parameter set of the multi-domain coupled reduced-order mechanism model and the boundary conditions of the residual correction network to complete the online self-evolution update of the digital twin model.

[0007] As a preferred embodiment of the real-time mapping digital twin simulation method for all elements of a hydropower station described in this invention, the steps of weighted fusion of a pre-set multi-domain benchmark coupling matrix based on the current operating condition mode to generate a time-varying dynamic coupling operator for driving the multi-domain coupled reduced-order mechanism model to perform real-time inference calculations, and obtaining the mechanism prediction output value, are as follows: A reference coupling matrix library is constructed, which includes a stable power generation reference matrix, a load regulation reference matrix, a start-stop transition reference matrix, and a fault disturbance reference matrix. Each reference coupling matrix is ​​obtained by joint calibration through offline high-fidelity simulation and historical measured data. The real-time collected hydropower station operation data is input into a pre-trained operating condition classifier, which outputs the probability distribution vector of the current time belonging to four typical operating condition modes. Using the elements in the probability distribution vector as weighting coefficients, the four corresponding benchmark coupling matrices in the benchmark coupling matrix library are linearly superimposed to generate a time-varying dynamic coupling operator. The time-varying dynamic coupling operator is substituted as the state transition matrix into the state equation of the multi-domain coupled reduced-order mechanism model. Numerical integration is then performed in conjunction with the initial value of the unified state vector of all elements to obtain the mechanism prediction output value.

[0008] As a preferred embodiment of the real-time mapping digital twin simulation method for all elements of a hydropower station described in this invention, the step of using elements in the probability distribution vector as weighting coefficients to linearly superimpose the four corresponding benchmark coupling matrices in the benchmark coupling matrix library to generate a time-varying dynamic coupling operator includes the following steps: The probability distribution vector is defined to include a first probability component, a second probability component, a third probability component, and a fourth probability component, which correspond to four operating modes: stable power generation, load regulation, start-stop transition, and fault disturbance, respectively. Extract the stable generation reference matrix corresponding to the first probability component, the load regulation reference matrix corresponding to the second probability component, the start-stop transition reference matrix corresponding to the third probability component, and the fault disturbance reference matrix corresponding to the fourth probability component from the reference coupling matrix library, and calculate the time-varying dynamic coupling operator.

[0009] As a preferred embodiment of the real-time mapping digital twin simulation method for all elements of a hydropower station described in this invention, the following steps are taken: The original residual sequence is obtained by calculating the difference between the measured data after spatiotemporal alignment processing and the mechanism prediction output value. The unified state vector of all elements, the real-time control input, and the original residual sequence are then input into a residual correction network embedded with physical conservation constraints to obtain the simulation correction amount. The measured data and the mechanism prediction output values ​​are interpolated and aligned according to the timestamp, and the original residual sequence is obtained by subtracting them channel by channel. The unified state vector of all elements, the real-time control input, and the original residual sequence are concatenated into a joint input vector. The joint input vector is input to the physical constraint layer of the residual correction network. The physical constraint layer contains a mass conservation constraint branch and an energy conservation constraint branch. The mass conservation constraint branch extracts the flow-related components in the joint input vector and calculates the inflow and outflow deviations. The energy conservation constraint branch extracts the pressure, velocity, and temperature-related components in the joint input vector and calculates the power balance deviations. By using the inflow / outflow deviation and power balance deviation as Lagrange multiplier constraints, a constrained nonlinear mapping is performed on the joint input vector, resulting in a simulation correction value with the same output dimension as the mechanistic prediction output value.

[0010] As a preferred embodiment of the real-time mapping digital twin simulation method for all elements of a hydropower station described in this invention, the specific steps are as follows: Determining a dynamic confidence factor based on sensor confidence assessment results, and then using the dynamic confidence factor to weightedly fuse the mechanism prediction output value and the simulation correction value to obtain a high-fidelity twin real-time output value. For each type of sensor, a real-time reliability score is calculated based on the variance, drift rate, and consistency index among redundant sensors of the measured data within a preset time period. The real-time confidence scores of all sensors are grouped by measurement channel and the average is calculated to obtain the grouped confidence vector; Construct a dynamic confidence factor matrix using the elements in the grouped confidence vector as diagonal elements; The simulation correction value is weighted using a dynamic confidence factor matrix and then added element by element to the mechanism prediction output value to obtain the real-time output value of the high-fidelity twin.

[0011] As a preferred embodiment of the real-time mapping digital twin simulation method for all elements of a hydropower station described in this invention, the following steps are taken: For each performance parameter of the equipment to be identified, the real-time excitation signal and the system response signal are selectively weighted in the frequency domain according to the feature identification frequency band, and the input-output cross-spectral coherence function is calculated. The product of the cross-spectral coherence function and the weighted spectral energy is then exponentially weighted and integrated within a preset sliding time window to obtain the effective energy density for parameter identification. The upper and lower limits of the corresponding feature identification frequency band are determined based on the physical mechanism of the performance parameters of the device to be identified. Short-time Fourier transforms are performed on the real-time excitation signal and the system response signal respectively to extract the spectral components in the frequency band and set the out-of-band spectral components to zero, thus obtaining the band-limited excitation spectrum and the band-limited response spectrum. Calculate the squared cross-spectral density modulus of the band-limited excitation spectrum and the band-limited response spectrum, and divide it by the product of the self-spectral density of the band-limited excitation spectrum and the self-spectral density of the band-limited response spectrum to obtain the input-output cross-spectral coherence function. The instantaneous effective identification energy is obtained by multiplying the input-output cross-spectral coherence function with the self-spectral density of the band-limited excitation spectrum at frequency points and then summing them within the feature identification frequency band. The instantaneous effective energy is accumulated and integrated within a preset sliding time window with an exponential decay weight to obtain the parameter identification effective energy density.

[0012] As a preferred embodiment of the real-time mapping digital twin simulation method for all elements of a hydropower station described in this invention, the specific steps for accumulating and integrating the instantaneous effective identified energy within a preset sliding time window with an exponentially decaying weight to obtain the parameter-identified effective energy density are as follows: Set the length of the preset sliding time window to The exponential decay constant is set to be ; Before obtaining the current time The instantaneous effective identification energy sequence within each sampling point is denoted as the th sampling point. The instantaneous effective identification energy of each sampling point is Calculate parameters to identify the effective energy density.

[0013] Secondly, this invention provides a real-time mapping digital twin simulation system for all elements of a hydropower station, comprising: The module includes a working condition identification module, a coupled inference module, a residual correction module, a confidence fusion module, a frequency domain identification module, and an adaptive update module. The working condition identification module is used to construct a unified state vector of all elements and identify the current working condition mode; The coupled inference module is used to generate a time-varying dynamic coupled operator by weighted fusion of the reference coupling matrix and obtain the mechanism prediction output value; The residual correction module is used to calculate the simulation correction amount based on the original residual sequence and the physical conservation constraint network; The confidence fusion module is used to predict the output value and the simulation correction amount using the dynamic confidence factor weighted fusion mechanism to obtain the real-time output value of the high-fidelity twin; The frequency domain identification module is used to identify the effective energy density by calculating parameters through frequency domain selective weighting and exponential weighted integration; The adaptive update module is used to identify the estimated values ​​of time-varying parameters for effective energy density inversion based on parameters and to complete the online self-evolution update of the model.

[0014] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the real-time mapping digital twin simulation method for all elements of a hydropower station as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the real-time mapping digital twin simulation method for all elements of a hydropower station as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are: By constructing a unified state vector for all elements and generating a time-varying dynamic coupling operator based on the probability-weighted fusion of operating conditions, a smooth and continuous derivation of the multi-domain coupled reduced-order mechanism model under complex mixed operating conditions is achieved, effectively eliminating the numerical oscillations caused by traditional hard switching. Combining a residual correction network embedded with physical conservation constraints and a dynamic confidence adaptive fusion mechanism based on the real-time reliability of sensors, the system's robustness to sensor anomalies is significantly improved while ensuring simulation accuracy. The effective energy density of parameter identification is obtained by using frequency-domain selective weighting and exponential decay integral, and this is used to modulate regularized recursive estimation to achieve online adaptive inversion of equipment performance parameters and model self-evolution update. This ensures that the digital twin can continuously track the physical state evolution of the real hydropower station, thus taking into account the computational efficiency, physical consistency, anti-interference ability, and long-term evolution accuracy of real-time simulation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the real-time mapping digital twin simulation method for all elements of a hydropower station in Example 1.

[0019] Figure 2 This is a schematic diagram of the real-time mapping digital twin simulation system for all elements of a hydropower station in Example 1. Detailed Implementation

[0020] To make the above-mentioned objectives, features and advantages of the present invention more readily understood, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, referring to Figure 1 and Figure 2 As one embodiment of the present invention, this embodiment provides a real-time mapping digital twin simulation method for all elements of a hydropower station, comprising the following steps: S1. Construct a unified state vector that includes all elements such as hydraulic, mechanical, electrical and control state variables, and identify the current operating mode based on real-time collected hydropower station operation data.

[0024] Furthermore, the unified state vector of all elements is defined as being composed of hydraulic sub-vectors, mechanical sub-vectors, electrical sub-vectors, and control sub-vectors concatenated in sequence. The hydraulic sub-vector includes upstream water level, downstream water level, water diversion flow rate, and volute pressure; the mechanical sub-vector includes guide vane opening, runner speed, main shaft swing, and bearing temperature; the electrical sub-vector includes stator voltage, stator current, active power, and reactive power; and the control sub-vector includes governor command, excitation voltage, and circuit breaker status.

[0025] Outlier removal and normalization are performed on the real-time collected hydropower station operation data. The processed data is then filled into the corresponding positions of the unified state vector of all elements according to the preset mapping relationship, completing the real-time update of the unified state vector of all elements. Active power, runner speed, guide vane opening and spiral casing pressure are extracted from the unified state vector of all elements as operating condition feature inputs, and the operating condition feature inputs are input into a pre-trained operating condition classifier. The pre-trained operating condition classifier outputs a probability distribution vector containing the probability of stable power generation, load regulation, start-stop transition and fault disturbance. The operating condition mode corresponding to the component with the largest value in the probability distribution vector is selected as the current operating condition mode recognition result.

[0026] It should be noted that by constructing a unified state vector covering multiple dimensions of hydraulic machinery, electrical systems, and control, and implementing data cleaning and standardized mapping, the dimensional differences and temporal misalignments between heterogeneous data can be eliminated, providing a consistent and reliable input foundation for subsequent multi-domain coupled modeling. At the same time, the output probability distribution of the operating condition classifier driven by key features, rather than discrete labels, enables the system to accurately capture the continuous and gradual changes in the operating state of the hydropower station, avoiding misjudgments and fluctuations caused by traditional threshold discrimination methods in the operating condition transition range, thereby improving the robustness and adaptability of full-element state perception and operating condition identification.

[0027] In particular, the design fundamentally solves the problem of model input distortion caused by heterogeneous data with different dimensions and time misalignment in traditional methods by constructing a unified state perception foundation covering multiple dimensions of hydraulics, mechanics, electrical, and control. Its advantage lies in using probability distributions rather than discrete hard labels to identify operating conditions. This enables the system to accurately capture the dynamic characteristics of the hydropower station that change continuously and gradually between different operating modes, effectively avoiding the misjudgments and state fluctuations that are prone to occur in the transition range of operating conditions in traditional threshold discrimination methods. This provides a highly robust and adaptable operating condition perception capability for subsequent digital twin simulations.

[0028] S2. Based on the current operating mode, the preset multi-domain benchmark coupling matrix is ​​weighted and fused to generate a time-varying dynamic coupling operator used to drive the multi-domain coupled reduced-order mechanism model to perform real-time inference operations, and the mechanism prediction output value is obtained.

[0029] Furthermore, a benchmark coupling matrix library is constructed, which includes a stable power generation benchmark matrix, a load regulation benchmark matrix, a start-stop transition benchmark matrix, and a fault disturbance benchmark matrix. Each benchmark coupling matrix is ​​obtained through joint calibration using offline high-fidelity simulation and historical measured data. Real-time collected hydropower station operation data is input into a pre-trained operating condition classifier, which outputs the probability distribution vector of the current time belonging to four typical operating condition modes.

[0030] Using the elements in the probability distribution vector as weighting coefficients, the four corresponding benchmark coupling matrices in the benchmark coupling matrix library are linearly superimposed to generate a time-varying dynamic coupling operator.

[0031] The probability distribution vector is defined to include a first probability component, a second probability component, a third probability component, and a fourth probability component, corresponding to four operating conditions: stable power generation, load regulation, start-stop transition, and fault disturbance, respectively. The stable power generation reference matrix corresponding to the first probability component, the load regulation reference matrix corresponding to the second probability component, the start-stop transition reference matrix corresponding to the third probability component, and the fault disturbance reference matrix corresponding to the fourth probability component are extracted from the reference coupling matrix library. The time-varying dynamic coupling operator is then calculated. ; In the formula, This represents a time-varying dynamic coupling operator. , , , These represent the first probability component, the second probability component, the third probability component, and the fourth probability component, respectively. , , , These represent the stable power generation reference matrix, load regulation reference matrix, start-stop transition reference matrix, and fault disturbance reference matrix, respectively.

[0032] In practice, hydropower stations are often in a mixed superposition of multiple ideal operating conditions rather than a single discrete state. By using a probability-weighted continuous interpolation method to replace hard switching logic, the model abrupt changes and numerical oscillations at the operating condition boundary can be eliminated, and the time-varying dynamic coupling operator can smoothly reflect the gradual characteristics of the real physical system, thereby ensuring the continuity and stability of the subsequent mechanism prediction output value.

[0033] The time-varying dynamic coupling operator is substituted as the state transition matrix into the state equation of the multi-domain coupled reduced-order mechanism model. Numerical integration is then performed in conjunction with the initial value of the unified state vector of all elements to obtain the mechanism prediction output value.

[0034] It should be noted that by obtaining a baseline coupling matrix library covering typical working conditions through offline calibration and generating time-varying dynamic coupling operators by linear superposition using real-time working condition probabilities as weights, the multi-domain coupling model can achieve smooth adaptive matching of complex mixed working conditions without increasing the online computing burden. This effectively avoids the numerical oscillations and simulation distortions caused by parameter mutations at the working condition switching boundary in traditional segmented modeling, ensuring that the mechanism prediction output maintains physical consistency and numerical stability throughout the entire operating envelope, and laying a solid model foundation for the real-time inference of high-fidelity digital twins.

[0035] In particular, the reference coupling matrix is ​​continuously interpolated and fused using a probability-weighted approach. The underlying logic is to acknowledge and adapt to the fact that the actual operation of a hydropower station is often a mixture of multiple ideal operating conditions. The biggest advantage of the design is that it completely abandons the hard switching logic at the operating condition switching boundary of traditional segmented modeling, eliminating the sudden changes in model parameters and numerical oscillations caused by this. This allows the generated time-varying dynamic coupling operator to smoothly and continuously reflect the gradual characteristics of the real system. Without increasing the online computing burden, it ensures the physical consistency and numerical stability of the mechanism prediction output throughout the entire operating envelope.

[0036] S3. Calculate the difference between the measured data after spatiotemporal alignment and the mechanism prediction output value to obtain the original residual sequence. Then, input the unified state vector of all elements, the real-time control input, and the original residual sequence into the residual correction network embedded with physical conservation constraints to obtain the simulation correction amount.

[0037] Furthermore, the measured data and the mechanism prediction output values ​​are interpolated and aligned according to timestamps, and the original residual sequence is obtained by subtracting each channel. The unified state vector of all elements, the real-time control input, and the original residual sequence are concatenated into a joint input vector. The joint input vector is then input into the physical constraint layer of the residual correction network. The physical constraint layer includes a mass conservation constraint branch and an energy conservation constraint branch. The mass conservation constraint branch extracts the flow-related components in the joint input vector and calculates the inflow and outflow deviations. The energy conservation constraint branch extracts the pressure, velocity, and temperature-related components in the joint input vector and calculates the power balance deviation. The inflow and outflow deviations and the power balance deviation are used as Lagrange multiplier constraint terms to perform a constrained nonlinear mapping on the joint input vector, outputting a simulation correction value with the same dimension as the mechanism prediction output value.

[0038] It should be noted that by spatiotemporally aligning the measured data with the predicted mechanistic values ​​and extracting the original residuals, and then jointly feeding the full-factor states, control inputs, and residual sequences into a correction network embedded with mass and energy conservation constraints, it is possible to forcibly impose physical law constraints while preserving the data-driven fitting ability. This prevents the pure black-box model from generating false corrections that violate physical laws under sparse samples or abnormal perturbations. The simulation correction quantity has both the ability to compensate for unmodeled dynamics and strictly satisfies the inherent conservation relationship of the system, thereby improving the reliability and generalization performance of the digital twin under non-ideal working conditions.

[0039] In particular, embedding physical conservation laws as constraints into the data-driven residual correction network is essentially designed to compensate for the blind fitting defects of pure black-box models when lacking physical understanding. The mechanism can forcibly constrain the output of the neural network, ensuring that it strictly adheres to basic physical rules such as the conservation of mass and energy when fitting unmodeled dynamics and measuring residuals. Its advantage lies in its ability to effectively prevent the model from generating false corrections that violate physical common sense, even when training samples are sparse or the system encounters abnormal perturbations. This significantly improves the credibility, generalization performance, and physical interpretability of the digital twin under non-ideal conditions.

[0040] S4. Based on the sensor confidence evaluation results, determine the dynamic confidence factor, and use the dynamic confidence factor to weight and fuse the mechanism prediction output value and the simulation correction value to obtain the high-fidelity twin real-time output value.

[0041] Furthermore, for each type of sensor, a real-time reliability score is calculated based on the variance, drift rate, and consistency index among redundant sensors of the measured data within a preset time period. The real-time reliability scores of all sensors are grouped by measurement channel and averaged to obtain a grouped confidence vector. The elements in the grouped confidence vector are used as diagonal elements to construct a dynamic confidence factor matrix. The simulation correction is weighted using the dynamic confidence factor matrix and then added element by element to the mechanism prediction output value to obtain the high-fidelity twin real-time output value.

[0042] Among them, the reliability of different sensors varies significantly under different operating conditions. If a uniform confidence level is applied to all correction values, the noise of low confidence channels will be amplified and pollute the overall output. By using a dynamic confidence factor matrix to achieve channel-level adaptive weighting, the real-time output value of the high-fidelity twin can fully absorb correction information when the sensor is healthy and automatically revert to the mechanism prediction value when the sensor is abnormal, thus balancing accuracy and robustness.

[0043] It should be noted that by comprehensively evaluating the statistical characteristics and redundancy consistency of historical sensor data to construct a channel-level dynamic confidence factor matrix, and using this matrix to adaptively weight and fuse simulation corrections, the contribution weight of each measurement channel to the final output can be intelligently adjusted according to the real-time health status of each measurement channel. When the sensor is normal, it fully absorbs measured information to improve accuracy, and when the sensor drifts or fails, it automatically suppresses noise interference and reverts to the mechanism prediction value, thereby achieving a dynamic balance between accuracy and robustness and effectively avoiding the risk of the entire twin output collapsing due to a single sensor failure.

[0044] In particular, a dynamic confidence factor based on real-time sensor health status assessment is introduced. Its design logic is to establish an adaptive risk-benefit balance mechanism. The design abandons the traditional approach of treating all measurement channels equally. It can intelligently adjust the contribution weight of the correction amount to the final output according to the real-time statistical characteristics and redundancy consistency of each sensor channel. Its advantage is that when the sensor is working normally, the system can fully absorb the measured information to improve the mapping accuracy of the twin. When the sensor drifts or fails, the system can automatically suppress the noise interference of that channel and smoothly fall back to the mechanism prediction value, thereby achieving a dynamic balance between accuracy and robustness, and effectively avoiding the risk of the entire twin output collapsing due to a single-point sensor failure.

[0045] S5. For each performance parameter of the device to be identified, the real-time excitation signal and the system response signal are selectively weighted in the frequency domain according to the feature identification frequency band, and the input-output cross-spectral coherence function is calculated. The product of the cross-spectral coherence function and the weighted spectral energy is exponentially weighted and integrated within a preset sliding time window to obtain the effective energy density for parameter identification.

[0046] Furthermore, based on the physical mechanism of the performance parameters of the device to be identified, the upper and lower limits of the corresponding feature identification frequency band are determined; short-time Fourier transforms are performed on the real-time excitation signal and the system response signal respectively to extract the spectral components within the feature identification frequency band and set the out-of-band spectral components to zero, thus obtaining the band-limited excitation spectrum and the band-limited response spectrum; the square of the cross-spectral density modulus of the band-limited excitation spectrum and the band-limited response spectrum is calculated and divided by the product of the self-spectral density of the band-limited excitation spectrum and the self-spectral density of the band-limited response spectrum to obtain the input-output cross-spectral coherence function; the input-output cross-spectral coherence function and the self-spectral density of the band-limited excitation spectrum are multiplied at each frequency point and then summed within the feature identification frequency band to obtain the instantaneous effective identification energy; the instantaneous effective identification energy is accumulated and integrated within a preset sliding time window with exponential decay weights to obtain the parameter identification effective energy density.

[0047] Set the length of the preset sliding time window to The exponential decay constant is set to be ; Get the current time before The instantaneous effective identification energy sequence within each sampling point is denoted as the th sampling point. The instantaneous effective identification energy of each sampling point is ; Calculate parameters to identify effective energy density: ; In the formula, The parameters indicate the effective energy density. Indicates the number of times since the current time. The instantaneous effective energy identification of each sampling point Represents the exponential decay constant. This indicates the preset sliding time window length.

[0048] The effectiveness of parameter identification depends not only on the signal strength at the current instant, but also on the continuous and causally clear accumulation of excitation over a period of time. By adopting exponential decay weighting instead of rectangular window averaging, the dominant role of recent effective excitation is preserved, while historical information is gradually forgotten. This allows the effective energy density of parameter identification to respond quickly to changes in operating conditions, while suppressing false triggering caused by transient spurious excitation. This provides a smooth and physically interpretable modulation basis for subsequent continuous adaptive inversion.

[0049] It should be noted that by defining characteristic identification frequency bands based on the physical mechanism of the equipment and combining cross-spectral coherence function and exponential decay weighted integral to calculate parameters and identify effective energy density, it is possible to accurately extract causal excitation components strongly correlated with target parameters from broadband noise. At the same time, more weight is given to recent effective signals while gradually forgetting old or transient interference information. This allows the obtained energy density index to not only sensitively reflect the true level of parameter identifiability under the current operating conditions, but also smoothly suppress false triggering caused by spurious excitation. This provides a physically interpretable and time-stable modulation basis for subsequent adaptive inversion, ensuring the reliability and convergence of the parameter identification process.

[0050] In particular, the effective energy density for parameter identification is calculated by defining the feature identification frequency band and combining it with the cross-spectral coherence function. The effectiveness of parameter identification depends not only on signal strength but also on continuous excitation with a clear causal relationship. An exponential decay weighting method is used to give higher weight to recent effective excitations while gradually invalidating old or transient interference information. The design advantage is that it can accurately extract the excitation components that are strongly correlated with the target parameters from the background noise of a wide frequency band, smoothly suppress false triggering caused by spurious excitations, and provide a modulation basis with clear physical meaning and stable timing for subsequent adaptive inversion, thus ensuring the reliability and convergence of the parameter identification process.

[0051] S6. Using the effective energy density identified by parameters as the modulation factor, a regularized recursive estimation algorithm is used to continuously and adaptively invert the performance parameters of the device to be identified to obtain time-varying parameter estimates. The time-varying parameter estimates are then injected into the physical parameter set of the multi-domain coupled reduced-order mechanism model and the boundary conditions of the residual correction network to complete the online self-evolution update of the digital twin model.

[0052] Furthermore, initial estimates of the performance parameters of the device to be identified, a regularization attenuation factor, and an initial covariance matrix are set, and the effective energy density of parameter identification is introduced as an adaptive gain modulation coefficient into the gain calculation stage of the regularization recursive estimation algorithm; the system observation output at the current moment and the model prediction output calculated by the multi-domain coupling reduction mechanism model based on the parameter estimates at the previous moment are obtained, and the difference between the system observation output and the model prediction output is calculated to obtain the parameter identification residual.

[0053] The regularized attenuation factor is dynamically scaled by the effective energy density identified by the parameter identification. The covariance matrix and gain matrix are iteratively updated by combining the initial covariance matrix and the parameter identification residual. The time-varying parameter estimate is obtained by correcting the initial estimate of the performance parameters of the device to be identified based on the gain matrix and the parameter identification residual.

[0054] The time-varying parameter estimates replace the original nominal parameter values ​​corresponding to the physical parameter set of the multi-domain coupled reduced-order mechanism model. At the same time, the time-varying parameter estimates are used as the boundary condition input port update values ​​of the physical constraint layer of the residual correction network. This allows the state transition characteristics of the multi-domain coupled reduced-order mechanism model and the constraint range of the residual correction network to synchronously follow the evolution of the real device performance, thus completing the online self-evolution update of the digital twin model.

[0055] It should be noted that by using the effective energy density identified by the parameter identification as a modulation factor to dynamically adjust the gain and attenuation characteristics of the regularized recursive estimation, and by synchronously injecting the time-varying parameters obtained from the inversion into the parameter set of the mechanism model and the boundary conditions of the residual correction network, the digital twin can continuously track the natural aging or sudden changes in the device performance during operation, achieving dual online self-evolution of the model structure and parameters. This avoids the accuracy degradation of the fixed parameter model after long-term operation and prevents parameter drift instability caused by unconstrained adaptive updates, thereby ensuring that the twin maintains a high degree of synchronization and mapping fidelity with the real physical system throughout its entire life cycle.

[0056] In particular, the effective energy density for parameter identification is used as a modulation factor to dynamically adjust the gain of the regularized recursive estimation and to simultaneously update the parameters of the mechanistic model and the residual correction network. The design essence is to endow the digital twin with an online self-evolving vitality. The mechanism enables the twin to continuously and stably track the natural aging or sudden changes in equipment performance during operation, achieving dual online calibration of model structure and parameters. Its advantages are that it effectively avoids the accuracy degradation of fixed parameter models due to changes in equipment status after long-term operation, and prevents parameter drift instability that may be caused by unconstrained adaptive updates through energy density modulation, ensuring that the twin maintains a high degree of synchronization and mapping fidelity with the real physical system throughout its entire life cycle.

[0057] This embodiment also provides a real-time mapping digital twin simulation system for all elements of a hydropower station, including: The module includes a working condition identification module, a coupled inference module, a residual correction module, a confidence fusion module, a frequency domain identification module, and an adaptive update module.

[0058] The working condition identification module is used to construct a unified state vector for all elements and identify the current working condition mode.

[0059] The coupled inference module is used to generate time-varying dynamic coupled operators by weighted fusion of the benchmark coupling matrix and obtain mechanism prediction output values.

[0060] The residual correction module is used to calculate the simulation correction amount based on the original residual sequence and the physical conservation constraint network.

[0061] The confidence fusion module is used to obtain the real-time output value of the high-fidelity twin by using the dynamic confidence factor weighted fusion mechanism to predict the output value and the simulation correction amount.

[0062] The frequency domain identification module is used to identify the effective energy density by calculating parameters through frequency domain selective weighting and exponential weighted integration.

[0063] The adaptive update module is used to identify the time-varying parameter estimates of the effective energy density inversion based on the parameters and to complete the online self-evolution update of the model.

[0064] This embodiment also provides a computer device applicable to the real-time mapping digital twin simulation method for all elements of a hydropower station, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the real-time mapping digital twin simulation method for all elements of a hydropower station as proposed in the above embodiment.

[0065] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0066] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the method for real-time mapping of all elements of a hydropower station using digital twin simulation as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0067] In summary, this invention constructs a unified state vector for all elements and generates a time-varying dynamic coupling operator based on probabilistic weighted fusion of operating conditions. This enables smooth and continuous derivation of a multi-domain coupled reduced-order mechanism model under complex mixed operating conditions, effectively eliminating numerical oscillations caused by traditional hard switching. By combining a residual correction network embedded with physical conservation constraints with a dynamic confidence adaptive fusion mechanism based on real-time sensor reliability, the system's robustness to sensor anomalies is significantly improved while ensuring simulation accuracy. The effective energy density for parameter identification is obtained by using frequency-domain selective weighting and exponential decay integral, and this is used to modulate regularized recursive estimation to achieve online adaptive inversion of equipment performance parameters and model self-evolution update. This ensures that the digital twin can continuously track the physical state evolution of the real hydropower station, thus balancing the computational efficiency, physical consistency, anti-interference capability, and long-term evolution accuracy of real-time simulation.

[0068] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A real-time mapping digital twin simulation method for all elements of a hydropower station, characterized in that: include, Construct a unified state vector that includes hydraulic, mechanical, electrical, and control state variables, and identify the current operating mode based on real-time collected hydropower station operation data; Based on the current operating conditions, the pre-set multi-domain benchmark coupling matrix is ​​weighted and fused to generate a time-varying dynamic coupling operator for driving the multi-domain coupled reduced-order mechanism model to perform real-time inference operations, and the mechanism prediction output value is obtained. The difference between the measured data after spatiotemporal alignment and the mechanism prediction output is calculated to obtain the original residual sequence. The unified state vector of all elements, the real-time control input and the original residual sequence are then input into the residual correction network embedded with physical conservation constraints to obtain the simulation correction. Based on the sensor confidence evaluation results, a dynamic confidence factor is determined, and the dynamic confidence factor is used to weight and fuse the mechanism prediction output value and the simulation correction value to obtain the real-time output value of the high-fidelity twin. For each device performance parameter to be identified, the real-time excitation signal and the system response signal are selectively weighted in the frequency domain according to the feature identification frequency band, and the input-output cross-spectral coherence function is calculated. The product of the cross-spectral coherence function and the weighted spectral energy is integrally weighted exponentially within a preset sliding time window to obtain the effective energy density for parameter identification. Using the effective energy density identified by parameters as the modulation factor, a regularized recursive estimation algorithm is used to continuously and adaptively invert the performance parameters of the device to be identified to obtain time-varying parameter estimates. The time-varying parameter estimates are then injected into the physical parameter set of the multi-domain coupled reduced-order mechanism model and the boundary conditions of the residual correction network to complete the online self-evolution update of the digital twin model.

2. The real-time mapping digital twin simulation method for all elements of a hydropower station as described in claim 1, characterized in that: The steps are as follows: Weighted fusion of the pre-set multi-domain baseline coupling matrix based on the current operating condition mode to generate a time-varying dynamic coupling operator for driving the multi-domain coupled reduced-order mechanism model to perform real-time inference operations, thereby obtaining the mechanism prediction output value. A reference coupling matrix library is constructed, which includes a stable power generation reference matrix, a load regulation reference matrix, a start-stop transition reference matrix, and a fault disturbance reference matrix. Each reference coupling matrix is ​​obtained by joint calibration through offline high-fidelity simulation and historical measured data. The real-time collected hydropower station operation data is input into a pre-trained operating condition classifier, which outputs the probability distribution vector of the current time belonging to four typical operating condition modes. Using the elements in the probability distribution vector as weighting coefficients, the four corresponding benchmark coupling matrices in the benchmark coupling matrix library are linearly superimposed to generate a time-varying dynamic coupling operator. The time-varying dynamic coupling operator is substituted as the state transition matrix into the state equation of the multi-domain coupled reduced-order mechanism model. Numerical integration is then performed in conjunction with the initial value of the unified state vector of all elements to obtain the mechanism prediction output value.

3. The real-time mapping digital twin simulation method for all elements of a hydropower station as described in claim 2, characterized in that: The step of using elements in the probability distribution vector as weighting coefficients to linearly superimpose the four corresponding benchmark coupling matrices in the benchmark coupling matrix library to generate a time-varying dynamic coupling operator is as follows: The probability distribution vector is defined to include a first probability component, a second probability component, a third probability component, and a fourth probability component, which correspond to four operating modes: stable power generation, load regulation, start-stop transition, and fault disturbance, respectively. Extract the stable generation reference matrix corresponding to the first probability component, the load regulation reference matrix corresponding to the second probability component, the start-stop transition reference matrix corresponding to the third probability component, and the fault disturbance reference matrix corresponding to the fourth probability component from the reference coupling matrix library, and calculate the time-varying dynamic coupling operator.

4. The real-time mapping digital twin simulation method for all elements of a hydropower station as described in claim 3, characterized in that: The original residual sequence is obtained by calculating the difference between the spatiotemporally aligned measured data and the mechanism prediction output. The unified state vector of all elements, the real-time control input, and the original residual sequence are then input into a residual correction network embedded with physical conservation constraints to obtain the simulation correction. The specific steps are as follows: The measured data and the mechanism prediction output values ​​are interpolated and aligned according to the timestamp, and the original residual sequence is obtained by subtracting them channel by channel. The unified state vector of all elements, the real-time control input, and the original residual sequence are concatenated into a joint input vector. The joint input vector is input to the physical constraint layer of the residual correction network. The physical constraint layer contains a mass conservation constraint branch and an energy conservation constraint branch. The mass conservation constraint branch extracts the flow-related components in the joint input vector and calculates the inflow and outflow deviations. The energy conservation constraint branch extracts the pressure, velocity, and temperature-related components in the joint input vector and calculates the power balance deviations. By using the inflow / outflow deviation and power balance deviation as Lagrange multiplier constraints, a constrained nonlinear mapping is performed on the joint input vector, resulting in a simulation correction value with the same output dimension as the mechanistic prediction output value.

5. The real-time mapping digital twin simulation method for all elements of a hydropower station as described in claim 4, characterized in that: The steps are as follows: First, a dynamic confidence factor is determined based on the sensor confidence assessment results. Then, the dynamic confidence factor is used to weight and fuse the mechanism prediction output value and the simulation correction value to obtain the high-fidelity twin's real-time output value. For each type of sensor, a real-time reliability score is calculated based on the variance, drift rate, and consistency index among redundant sensors of the measured data within a preset time period. The real-time confidence scores of all sensors are grouped by measurement channel and the average is calculated to obtain the grouped confidence vector; Construct a dynamic confidence factor matrix using the elements in the grouped confidence vector as diagonal elements; The simulation correction value is weighted using a dynamic confidence factor matrix and then added element by element to the mechanism prediction output value to obtain the real-time output value of the high-fidelity twin.

6. The real-time mapping digital twin simulation method for all elements of a hydropower station as described in claim 5, characterized in that: For each performance parameter of the device to be identified, the real-time excitation signal and the system response signal are selectively weighted in the frequency domain according to the feature identification frequency band, and the input-output cross-spectral coherence function is calculated. The product of the cross-spectral coherence function and the weighted spectral energy is then exponentially weighted and integrated within a preset sliding time window to obtain the effective energy density for parameter identification. The specific steps are as follows: The upper and lower limits of the corresponding feature identification frequency band are determined based on the physical mechanism of the performance parameters of the device to be identified. Short-time Fourier transforms are performed on the real-time excitation signal and the system response signal respectively to extract the spectral components in the frequency band and set the out-of-band spectral components to zero, thus obtaining the band-limited excitation spectrum and the band-limited response spectrum. Calculate the squared cross-spectral density modulus of the band-limited excitation spectrum and the band-limited response spectrum, and divide it by the product of the self-spectral density of the band-limited excitation spectrum and the self-spectral density of the band-limited response spectrum to obtain the input-output cross-spectral coherence function. The instantaneous effective identification energy is obtained by multiplying the input-output cross-spectral coherence function with the self-spectral density of the band-limited excitation spectrum at frequency points and then summing them within the feature identification frequency band. The instantaneous effective energy is accumulated and integrated within a preset sliding time window with an exponential decay weight to obtain the parameter identification effective energy density.

7. The real-time mapping digital twin simulation method for all elements of a hydropower station as described in claim 6, characterized in that: The specific steps for accumulating and integrating the instantaneous effective identification energy within a preset sliding time window with an exponentially decaying weight to obtain the parameter identification effective energy density are as follows: Set the length of the preset sliding time window to The exponential decay constant is set to be ; Before obtaining the current time The instantaneous effective identification energy sequence within each sampling point is denoted as the th sampling point. The instantaneous effective identification energy of each sampling point is And calculate parameters to identify the effective energy density.

8. A real-time mapping digital twin simulation system for all elements of a hydropower station, based on the real-time mapping digital twin simulation method for all elements of a hydropower station as described in any one of claims 1 to 7, characterized in that: include: The module includes a working condition identification module, a coupled inference module, a residual correction module, a confidence fusion module, a frequency domain identification module, and an adaptive update module. The working condition identification module is used to construct a unified state vector of all elements and identify the current working condition mode; The coupled inference module is used to generate a time-varying dynamic coupled operator by weighted fusion of the reference coupling matrix and obtain the mechanism prediction output value; The residual correction module is used to calculate the simulation correction amount based on the original residual sequence and the physical conservation constraint network; The confidence fusion module is used to predict the output value and the simulation correction amount using the dynamic confidence factor weighted fusion mechanism to obtain the real-time output value of the high-fidelity twin; The frequency domain identification module is used to identify the effective energy density by calculating parameters through frequency domain selective weighting and exponential weighted integration; The adaptive update module is used to identify the estimated values ​​of time-varying parameters for effective energy density inversion based on parameters and to complete the online self-evolution update of the model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the real-time mapping digital twin simulation method for all elements of a hydropower station as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the real-time mapping digital twin simulation method for all elements of a hydropower station as described in any one of claims 1 to 7.

Images