Water-turbine generator set deformation tracking prediction method and system based on model reconstruction

Abstract

The invention relates to the technical field of hydroelectric generating set deformation prediction, and provides a hydroelectric generating set deformation tracking prediction method and system based on model reconstruction, and the method comprises the steps: collecting multi-source monitoring data; constructing a static deformation basic model, and establishing a multi-physics field coupling analysis model based on the static deformation basic model and the deformation influence factors; performing dynamic parameter identification and updating on the multi-physics field coupling analysis model to obtain a dynamic deformation model; preprocessing the multi-source monitoring data to obtain standardized data, inputting the standardized data into the dynamic deformation model, optimizing the output of the dynamic deformation model, and establishing a multi-scale prediction mechanism to obtain a deformation prediction result; and comparing a deformation prediction result with actually monitored deformation data, calculating a prediction error, performing adaptive optimization on model parameters of the dynamic deformation model, and outputting an optimized deformation prediction result. According to the invention, accurate prediction of the deformation of the water-turbine generator set is realized, and the accuracy of deformation prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of deformation prediction of hydro-generator sets, and particularly to a deformation tracking and prediction method and system for hydro-generator sets based on model reconstruction. Background Art

[0002] Hydro-generator sets are the core equipment of hydropower stations, and their operating status is directly related to the safety and efficiency of the power stations. During the operation of hydro-generator sets, they are affected by various physical factors such as water pressure, temperature changes, and vibrations, and are prone to radial and axial deformations. If these deformations are not monitored and warned in time, it may lead to a decrease in the operating efficiency of the unit, and in severe cases, it may even cause equipment failures. Especially at key nodes such as the upper support of the unit, the guide bearing position, the rotor and the stator, the deformation problem is more prominent. Traditional deformation monitoring of hydro-generator sets mainly relies on single monitoring means, such as using displacement sensors or vibration sensors alone for monitoring. These monitoring methods often have problems such as limited data acquisition dimensions and incomplete monitoring parameters.

[0003] The prior art usually uses fixed monitoring equipment for deformation monitoring of hydro-generator sets, and analyzes deformation data through simple data processing methods. Due to the lack of comprehensive consideration of the coupling effects of multiple physical fields such as water pressure, temperature, and vibration, it is difficult to accurately reflect the actual deformation state of hydro-generator sets. Various monitoring data are often processed separately, resulting in low correlation of monitoring information, lack of dynamic parameter identification and adaptive optimization capabilities, and model parameters cannot be updated in real time according to the actual monitoring effect, affecting the accuracy of deformation prediction. Due to the lack of a multi-scale prediction mechanism, it is difficult to simultaneously consider the prediction of short-term and long-term deformation trends, thus affecting the reliability of deformation prediction results. Summary of the Invention

[0004] In view of this, the present invention proposes a deformation tracking and prediction method and system for hydro-generator sets based on model reconstruction, which comprehensively considers the influences of water pressure, temperature, and vibration, and solves the problem that the prior art lacks a multi-scale prediction mechanism and is difficult to simultaneously consider the prediction of short-term and long-term deformation trends.

[0005] The technical solution of the present invention is implemented as follows: In the first aspect, the present invention provides a deformation tracking and prediction method for hydro-generator sets based on model reconstruction, including the following steps:

[0006] Arrange a sensor network at preset key positions of the hydro-generator set, and collect multi-source monitoring data of the hydro-generator set through the sensor network;

[0007] Use the finite element method to construct a static deformation basic model of the hydro-generator set, and establish a multi-physical field coupling analysis model of the hydro-generator set based on the static deformation basic model and deformation influencing factors;

[0008] The recursive least squares method is used to perform dynamic parameter identification and update on the multi-physical field coupling analysis model to obtain a dynamic deformation model;

[0009] The multi-source monitoring data is preprocessed to obtain standardized data, the standardized data is input into the dynamic deformation model, and a deep learning algorithm is used to optimize the output of the dynamic deformation model, and a multi-scale prediction mechanism for short-term and long-term deformation trends is established to obtain the deformation prediction result of the hydro-generator unit;

[0010] The deformation prediction result is compared with the actually monitored deformation data, the prediction error is calculated, the model parameters of the dynamic deformation model are adaptively optimized based on the prediction error, and the optimized deformation prediction result is output.

[0011] Based on the above technical solutions, preferably, a sensor network is arranged at preset key positions of the hydro-generator unit, and multi-source monitoring data of the hydro-generator unit is collected through the sensor network, specifically including:

[0012] Optical fiber sensors, acceleration sensors and temperature sensors are arranged at the upper support of the unit, guide bearing position, key nodes of the rotor and key nodes of the stator of the hydro-generator unit to construct a multi-source sensor network;

[0013] The optical fiber sensor is used to collect deformation data, including radial displacement and axial displacement; the acceleration sensor is used to collect vibration data, including radial vibration and axial vibration; the temperature sensor is used to collect temperature data, including stator temperature and rotor temperature; a pressure sensor is also included, which is used to collect water pressure data; a speed sensor is used to collect speed data;

[0014] Multi-source monitoring data is collected in real time through the multi-source sensor network, and the multi-source monitoring data is subjected to time series synchronization processing;

[0015] The time stamp alignment method is used to synchronize the multi-source monitoring data, the data collected by all sensors is resampled at a unified sampling frequency, and the missing data is supplemented by an interpolation algorithm.

[0016] Based on the above technical solutions, preferably, the finite element method is used to construct a static deformation basic model of the hydro-generator unit, and a multi-physical field coupling analysis model of the hydro-generator unit is established based on the static deformation basic model and deformation influencing factors, specifically including:

[0017] Based on the geometric parameters and material parameters of the hydro-generator unit, the finite element method is used to construct a static deformation basic model, and a mapping relationship between the deformation amount and the external load is established;

[0018] The tetrahedral elements are used to mesh the hydro-generator set, establish the stiffness equation of the nodal displacement and the external load, and establish the relationship between the deformation amount and the stress and strain through the strain energy density function; the deformation amount includes the radial deformation and the axial deformation, and the external load includes the gravity load, the water pressure load, the temperature load and the dynamic load;

[0019] Taking the water pressure load, the temperature field distribution and the vibration response as the boundary conditions, establish the multi-physical field coupling analysis model of the hydro-generator set;

[0020] Establish the coupling equations of the water pressure field, the temperature field and the vibration field, construct the multi-physical field coupling coefficient matrix by the block matrix method, and perform the multi-physical field coupling calculation through iterative solution.

[0021] On the basis of the above technical solutions, preferably, the recursive least squares method is used to identify and update the dynamic parameters of the multi-physical field coupling analysis model to obtain the dynamic deformation model, which specifically includes:

[0022] Based on the multi-source monitoring data, construct the parameter identification equation, and use the recursive least squares method to identify the model parameters of the multi-physical field coupling analysis model in real time to obtain the initial estimated values of the model parameters;

[0023] The parameter identification equation includes the observation equation and the state equation, and the recursive least squares method adopts the weighted least squares criterion;

[0024] According to the newly obtained monitoring data, recursively update the model parameters, and introduce a forgetting factor to dynamically adjust the weight of the historical data to achieve the adaptive optimization of the model parameters;

[0025] Introduce an exponentially decaying forgetting factor to dynamically adjust the weight of the historical data to make the model pay more attention to the recent data; evaluate the reliability of the parameter identification by calculating the covariance matrix of the parameter estimation, and dynamically adjust the value of the forgetting factor according to the reliability index.

[0026] On the basis of the above technical solutions, preferably, the calculation formulas of the observation equation and the state equation are:

[0027] y(k1) = H(k1)θ(k1) + v(k1);

[0028] θ(k1) = θ(k1 - 1) + w(k1);

[0029] Among them, y(k1) is the observation vector at time k1, H(k1) is the observation matrix at time k1, θ(k1) is the parameter vector to be identified at time k1, v(k1) is the observation noise vector at time k1, θ(k1) is the parameter vector at time k1, θ(k1 - 1) is the parameter vector at time k1 - 1, and w(k1) is the parameter evolution noise vector;

[0030] The calculation formula of the recursive least squares method is:

[0031]

[0032] K(k1) = P(k1 - 1)H T (k1)[λ(k1)I + H(k1)P(k1 - 1)H T (k1)] -1 ;

[0033] Among them, is the updated parameter vector at time k1, is the parameter vector at time k1 - 1, K(k1) is the gain matrix at time k1, is the residual vector, P(k1 - 1) is the covariance matrix at time k1 - 1, λ(k1) is the forgetting factor at time k1, and I is the identity matrix;

[0034] The calculation formula of the covariance matrix is:

[0035]

[0036] Among them, P(k1) is the covariance matrix at time k1, P(k1 - 1) is the covariance matrix at time k1 - 1, λ(k1) is the forgetting factor at time k1, is the dynamic forgetting factor, λ0 is the basic forgetting factor, β1 is the adjustment coefficient, e(k1) is the estimation error vector at time k1, and exp(·) is the exponential function;

[0037] The calculation formula of the reliability index is:

[0038]

[0039] Among them, R(k1) is the reliability index at time k1, tr[P(k1)] is the trace of the covariance matrix P(k1), and tr[P(0)] is the trace of the initial covariance matrix.

[0040] Based on the above technical solutions, preferably, the multi-source monitoring data is preprocessed to obtain standardized data, the standardized data is input into the dynamic deformation model, and a deep learning algorithm is used to optimize the output of the dynamic deformation model, and a multi-scale prediction mechanism for short-term and long-term deformation trends is established to obtain the deformation prediction result of the hydro-generator unit, which specifically includes:

[0041] The noise and outliers in the monitoring data are removed by a filtering method, and the monitoring data with different dimensions is converted to a unified numerical range by using the min-max normalization or Z-score normalization method to obtain standardized data;

[0042] The deep learning algorithm includes a convolutional neural network and a long short-term memory network. The convolutional neural network is used to extract the spatial features in the standardized monitoring data, and the long short-term memory network is used to model the time series dependence relationship of the deformation data.

[0043] Based on the above technical solutions, preferably, the deformation prediction result is compared with the actually monitored deformation data, the prediction error is calculated, and the model parameters of the dynamic deformation model are adaptively optimized based on the prediction error, and the optimized deformation prediction result is output, which specifically includes:

[0044] The weighted absolute error formula is used to calculate the prediction error between the predicted value and the actual value, and the calculation formula is:

[0045]

[0046] where WAE is the weighted absolute error, A is the total number of samples of the monitoring data, ω a is the weight coefficient of the a-th sample, y pred (a) is the predicted deformation value of the a-th sample, and y actual (a) is the actually monitored deformation value of the a-th sample;

[0047] The model parameters of the dynamic deformation model are iteratively updated by the gradient descent method, and the calculation formula is:

[0048]

[0049] where δ T+1 is the model parameter after the (T + 1)-th iteration, δ T is the model parameter after the T-th iteration, η is the learning rate, is the gradient of the model parameter δ, λ1 is the regularization coefficient, Δδ T is the adjustment amount of the model parameter in the T-th iteration, μ is the momentum coefficient, Δδ T-1 is the adjustment amount of the model parameter in the (T - 1)-th iteration, is the gradient of the error function with respect to the model parameter δ.

[0050] In a second aspect, the present invention also provides a deformation tracking and prediction system for a hydro-generator unit based on model reconstruction, and the system includes:

[0051] A sensor acquisition module, configured to arrange a sensor network at preset key positions of the hydro-generator unit, and acquire multi-source monitoring data of the hydro-generator unit through the sensor network;

[0052] A static deformation module, configured to construct a static deformation basic model of the hydro-generator unit by using the finite element method, and establish a multi-physical field coupling analysis model of the hydro-generator unit based on the static deformation basic model and deformation influencing factors;

[0053] A dynamic deformation module, configured to perform dynamic parameter identification and update on the multi-physical field coupling analysis model by using the recursive least squares method to obtain a dynamic deformation model;

[0054] A deformation prediction module, configured to preprocess the multi-source monitoring data to obtain standardized data, input the standardized data into the dynamic deformation model, optimize the output of the dynamic deformation model by using a deep learning algorithm, establish a multi-scale prediction mechanism for short-term and long-term deformation trends, and obtain a deformation prediction result of the hydro-generator unit;

[0055] An error optimization module, configured to compare the deformation prediction result with the actually monitored deformation data, calculate a prediction error, adaptively optimize the model parameters of the dynamic deformation model based on the prediction error, and output an optimized deformation prediction result.

[0056] In a third aspect, the present invention also provides an electronic device, including: at least one processor, at least one memory, a communication interface, and a bus;

[0057] Wherein, the processor, the memory, and the communication interface complete communication with each other through the bus, the memory stores program instructions executable by the processor, and the processor calls the program instructions to implement the steps of a deformation tracking and prediction method for a hydro-generator unit based on model reconstruction.

[0058] In a fourth aspect, the present invention also provides a computer-readable storage medium, and the computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the steps of a deformation tracking and prediction method for a hydro-generator unit based on model reconstruction.

[0059] A deformation tracking and prediction method and system for a hydro-generator unit based on model reconstruction according to the present invention have the following beneficial effects compared with the prior art:

[0060] (1) By arranging a multi-source sensor network at key positions of the hydro-generator unit, collecting multi-source monitoring data such as deformation, vibration, temperature, water pressure, and rotational speed, constructing a static deformation basic model by combining the finite element method, and performing dynamic parameter identification and update based on the multi-physical field coupling analysis model, a dynamic deformation model is established. The output of the dynamic deformation model is optimized using deep learning algorithms, a multi-scale prediction mechanism for short-term and long-term deformation trends is constructed, and the model parameters are dynamically adjusted through adaptive optimization, thereby achieving accurate prediction of the deformation of the hydro-generator unit and improving the comprehensiveness of monitoring and the accuracy of prediction;

[0061] (2) By constructing an observation equation and a state equation, combining the recursive least squares method to perform real-time identification and dynamic update of the parameters of the multi-physical field coupling analysis model, and introducing a forgetting factor to adjust the weights of historical data, the accuracy and adaptability of model parameter identification are improved;

[0062] (3) By comparing the deformation prediction results with the actual monitoring data, calculating the prediction error, and adaptively optimizing the parameters of the dynamic deformation model based on the error feedback, the dynamic adjustment and continuous improvement of the model are achieved, the prediction deviation can be effectively corrected, the accuracy and reliability of deformation prediction are improved, and at the same time, the adaptability of the model to complex operating conditions is enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0064] Figure 1 It is a flowchart of a method for tracking and predicting the deformation of a hydro-generator unit based on model reconstruction according to the present invention;

[0065] Figure 2 It is a structural diagram of a system for tracking and predicting the deformation of a hydro-generator unit based on model reconstruction according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in combination with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0067] Please refer to Figure 1, the present invention provides a deformation tracking and prediction method for a hydro-generator set based on model reconstruction, comprising the following steps:

[0068] Arrange a sensor network at preset key positions of the hydro-generator set, and collect multi-source monitoring data of the hydro-generator set through the sensor network;

[0069] Adopt the finite element method to construct a static deformation basic model of the hydro-generator set, and establish a multi-physical field coupling analysis model of the hydro-generator set based on the static deformation basic model and deformation influencing factors;

[0070] Adopt the recursive least squares method to perform dynamic parameter identification and update on the multi-physical field coupling analysis model to obtain a dynamic deformation model;

[0071] Preprocess the multi-source monitoring data to obtain standardized data, input the standardized data into the dynamic deformation model, and adopt a deep learning algorithm to optimize the output of the dynamic deformation model, establish a multi-scale prediction mechanism for short-term and long-term deformation trends, and obtain the deformation prediction result of the hydro-generator set;

[0072] Compare the deformation prediction result with the actually monitored deformation data, calculate the prediction error, adaptively optimize the model parameters of the dynamic deformation model based on the prediction error, and output the optimized deformation prediction result.

[0073] Specifically, in this embodiment, a multi-source sensor network is arranged at the key positions of the hydro-generator set to collect multi-source monitoring data such as deformation, vibration, temperature, water pressure, and rotational speed, a static deformation basic model is constructed in combination with the finite element method, dynamic parameter identification and update are performed based on the multi-physical field coupling analysis model to establish a dynamic deformation model, the output of the dynamic deformation model is optimized by using a deep learning algorithm, a multi-scale prediction mechanism for short-term and long-term deformation trends is constructed, and the model parameters are dynamically adjusted through adaptive optimization, thereby realizing the accurate prediction of the deformation of the hydro-generator set and improving the comprehensiveness of monitoring and the accuracy of prediction.

[0074] The arrangement of the sensor network at the preset key positions of the hydro-generator set and the collection of multi-source monitoring data of the hydro-generator set through the sensor network specifically include:

[0075] Arrange fiber optic sensors, acceleration sensors, and temperature sensors at the upper support of the hydro-generator set, the guide bearing position, the key nodes of the rotor, and the key nodes of the stator to construct a multi-source sensor network;

[0076] The optical fiber sensor is used to collect deformation data, including radial displacement and axial displacement; the acceleration sensor is used to collect vibration data, including radial vibration and axial vibration; the temperature sensor is used to collect temperature data, including stator temperature and rotor temperature; it also includes a pressure sensor for collecting water pressure data and a speed sensor for collecting speed data;

[0077] The multi-source sensor network is used to collect multi-source monitoring data in real time, and the multi-source monitoring data is subjected to time series synchronization processing to ensure the time series consistency of the data;

[0078] The time stamp alignment method is used to synchronize the multi-source monitoring data. The data collected by all sensors are resampled at a unified sampling frequency, and the missing data are supplemented by an interpolation algorithm to achieve the time series alignment of the multi-source monitoring data.

[0079] Specifically, in this embodiment, optical fiber sensors, acceleration sensors, temperature sensors, pressure sensors and speed sensors are arranged at the upper support of the hydro-generator unit, the guide bearing position, the key nodes of the rotor and the key nodes of the stator to construct a multi-source sensor network, realizing the comprehensive monitoring of the key parts of the hydro-generator unit. The optical fiber sensor can accurately collect radial displacement and axial displacement data, the acceleration sensor can capture radial vibration and axial vibration information, the temperature sensor can monitor the temperature changes of the stator and rotor, and the pressure sensor and speed sensor respectively provide water pressure and speed data.

[0080] By performing time series synchronization processing on the multi-source monitoring data, using the time stamp alignment method and the interpolation algorithm to resample and supplement the missing data, the time series consistency and integrity of the multi-source data are ensured, the relevance and accuracy of the data are improved, the comprehensive perception and accurate monitoring of the deformation state of the hydro-generator unit are realized, and the reliability of the monitoring system is enhanced.

[0081] The finite element method is used to construct a static deformation basic model of the hydro-generator unit, and a multi-physical field coupling analysis model of the hydro-generator unit is established based on the static deformation basic model and deformation influencing factors, specifically including:

[0082] Based on the geometric parameters and material parameters of the hydro-generator unit, the finite element method is used to construct a static deformation basic model and establish the mapping relationship between the deformation amount and the external load;

[0083] The tetrahedral element is used to mesh the hydro-generator unit, establish the stiffness equation of the node displacement and the external load, and establish the relationship between the deformation amount and the stress and strain through the strain energy density function; the deformation amount includes radial deformation and axial deformation, and the external load includes gravity load, water pressure load, temperature load and dynamic load;

[0084] The calculation formula of the stiffness equation is as follows:

[0085] [K]{u} = {F};

[0086] Where, [K] is the stiffness matrix, {u} is the nodal displacement vector, and {F} is the external load vector;

[0087] The calculation formula of the strain energy density function is as follows:

[0088]

[0089] Where, W1 is the strain energy density, ε1 is the strain vector, and D1 is the elastic matrix;

[0090] Taking the water pressure load, temperature field distribution, and vibration response as boundary conditions, a multi-physical field coupling analysis model of the hydro-generator unit is established to realize the conversion from static deformation to multi-physical field coupling analysis;

[0091] The coupling equations of the water pressure field, temperature field, and vibration field are established. The multi-physical field coupling coefficient matrix is constructed by the block matrix method, and the multi-physical field coupling calculation is carried out through iterative solution; the coupling calculation considers the action effect of water pressure on deformation, the thermal stress effect caused by temperature gradient, and the dynamic stress effect caused by vibration;

[0092] The calculation formula of the coupling equation is as follows:

[0093]

[0094] Where, [K 11 is the stiffness matrix of the deformation field, [K 22 is the heat conduction matrix of the temperature field, [K 33 is the mass and damping matrix of the vibration field, [K 12 is the thermal stress effect matrix of the temperature field on the deformation field, [K 13 is the vibration field on the deformation field, [K 21 is the thermo-mechanical coupling effect matrix of the deformation field on the temperature field, [K 23 is the vibration thermal effect matrix of the vibration field on the temperature field, [K 31 is the structural stiffness effect matrix of the deformation field on the vibration field, [K 32 is the influence matrix of the temperature field on the vibration field, {u1} is the displacement vector of the deformation field, {u2} is the temperature distribution vector of the temperature field, {u3} is the amplitude vector of the vibration field, {F1} is the external load vector of the deformation field, {F2} is the thermal load vector of the temperature field, and {F3} is the dynamic load vector of the vibration field;

[0095] The update formula of the multi-physical field coupling coefficient matrix is as follows:

[0096]

[0097] Among them, K t+1 is the coupling coefficient matrix at time t + 1, and K t is the coupling coefficient matrix at time t. α1 is the update coefficient, is the load, is the displacement.

[0098] Specifically, in this embodiment, by adopting the finite element method and combining the stiffness equation and the strain energy density function, a static deformation basic model of the hydro-generator unit is established. Further, based on boundary conditions such as water pressure load, temperature field distribution, and vibration response, a multi-physical field coupling analysis model is constructed. This model comprehensively considers the action effect of water pressure on deformation, the thermal stress effect caused by temperature gradient, and the dynamic stress effect caused by vibration, and realizes the conversion from static deformation to multi-physical field coupling analysis. Compared with the traditional single-physical field analysis method, this model can more comprehensively reflect the deformation state of the hydro-generator unit under complex operating conditions.

[0099] By establishing the coupling equations of the water pressure field, temperature field, and vibration field, and adopting the block matrix method to construct the multi-physical field coupling coefficient matrix, the coupling effect between multi-physical fields can be effectively modeled and calculated. The application of the block matrix method improves the efficiency and accuracy of the coupling calculation, and avoids the problem of excessive calculation overhead caused by complex coupling relationships in the traditional method.

[0100] In the coupling calculation, by iteratively solving the multi-physical field coupling equation, the comprehensive influence of water pressure, temperature, and vibration on deformation can be dynamically captured. Specifically, the model can accurately describe complex coupling relationships such as the thermo-mechanical coupling effect of the temperature field on the deformation field, the dynamic stress effect of the vibration field on the deformation field, and the structural stiffness effect of the deformation field on the vibration field.

[0101] The dynamic parameter identification and update of the multi-physical field coupling analysis model by using the recursive least squares method to obtain a dynamic deformation model specifically includes:

[0102] Constructing a parameter identification equation based on multi-source monitoring data, and using the recursive least squares method to identify the model parameters of the multi-physical field coupling analysis model in real time to obtain the initial estimated values of the model parameters;

[0103] The parameter identification equation includes an observation equation and a state equation. The observation equation describes the relationship between the monitoring data and the model parameters, and the state equation describes the dynamic change characteristics of the model parameters. The recursive least squares method adopts the weighted least squares criterion to achieve the optimal estimation of the parameters by minimizing the weighted residual sum of squares;

[0104] Based on the newly acquired monitoring data, the model parameters are recursively updated, and a forgetting factor is introduced to dynamically adjust the weights of historical data, realizing the adaptive optimization of the model parameters;

[0105] An exponentially decaying forgetting factor is introduced to dynamically adjust the weights of historical data, enabling the model to pay more attention to recent data; by calculating the covariance matrix of parameter estimation, the reliability of parameter identification is evaluated, and the value of the forgetting factor is dynamically adjusted according to the reliability index.

[0106] Specifically, in this embodiment, by constructing a parameter identification equation based on multi-source monitoring data and using the recursive least squares method to perform real-time identification of the model parameters of the multi-physical field coupling analysis model, the initial values of the model parameters can be estimated quickly and accurately. The recursive least squares method uses the weighted least squares criterion to achieve the optimal estimation of parameters by minimizing the weighted sum of squared residuals, ensuring the real-time nature of model parameter identification.

[0107] During the parameter identification process, by introducing a forgetting factor to dynamically adjust the weights of historical data, the model pays more attention to the changing characteristics of recent data, thereby improving the model's adaptability to dynamic working conditions. The exponential decay mechanism of the forgetting factor can effectively reduce the influence of historical data on the current parameter estimation.

[0108] By calculating the covariance matrix of parameter estimation, evaluating the reliability of parameter identification, and dynamically adjusting the value of the forgetting factor according to the reliability index, this embodiment can further optimize the parameter update process while ensuring the accuracy of model parameter identification, enhancing the robustness of the model.

[0109] The calculation formulas of the observation equation and the state equation are as follows:

[0110] y(k1) = H(k1)θ(k1) + v(k1);

[0111] θ(k1) = θ(k1 - 1) + w(k1);

[0112] where y(k1) is the observation vector at time k1, H(k1) is the observation matrix at time k1, θ(k1) is the parameter vector to be identified at time k1, v(k1) is the observation noise vector at time k1, θ(k1) is the parameter vector at time k1, θ(k1 - 1) is the parameter vector at time k1 - 1, and w(k1) is the parameter evolution noise vector;

[0113] The calculation formula of the recursive least squares method is as follows:

[0114]

[0115] K(k1) = P(k1 - 1)H T(k1)[λ(k1)I + H(k1)P(k1 - 1)H T (k1)] -1 ;

[0116] wherein, is the parameter vector at the updated k1 moment, is the parameter vector at the k1 - 1 moment, K(k1) is the gain matrix at the k1 moment, is the residual vector, P(k1 - 1) is the covariance matrix at the k1 - 1 moment, λ(k1) is the forgetting factor at the k1 moment, and I is the identity matrix;

[0117] The calculation formula of the covariance matrix is:

[0118]

[0119] wherein, P(k1) is the covariance matrix at the k1 moment, P(k1 - 1) is the covariance matrix at the k1 - 1 moment, λ(k1) is the forgetting factor at the k1 moment, is the dynamic forgetting factor, λ0 is the basic forgetting factor, β1 is the adjustment coefficient, e(k1) is the estimation error vector at the k1 moment, and exp(·) is the exponential function;

[0120] The calculation formula of the reliability index is:

[0121]

[0122] wherein, R(k1) is the reliability index at the k1 moment, tr[P(k1)] is the trace of the covariance matrix P(k1), and tr[P(0)] is the trace of the initial covariance matrix.

[0123] Specifically, in this embodiment, the relationship between the monitoring data and the model parameters is described by the observation equation, which can associate multi - source monitoring data with the parameters of the dynamic deformation model. The state equation is used to describe the dynamic change characteristics of the model parameters and can reflect the change law of the model parameters at different time steps. By combining the observation equation and the state equation, the model can simultaneously consider the real - time nature of the monitoring data and the dynamics of parameter changes.

[0124] By calculating the covariance matrix, the reliability of the model parameter estimation is quantified. The dynamic update mechanism of the covariance matrix can reflect the change of the accuracy of the model parameter identification in real time, ensuring the credibility of the model parameter estimation.

[0125] The recursive least squares method is used to solve the observation equation and the state equation, and the optimal estimation of the model parameters is achieved by minimizing the weighted residual sum of squares. The recursive least squares method can update the model parameters in real time after each new monitoring data is obtained, avoiding the problem of recalculation required by the traditional batch least squares method and greatly improving the calculation efficiency. Combining the forgetting factor mechanism, the recursive least squares method can dynamically adjust the weights of historical data, making the model pay more attention to the change characteristics of recent data, thus enhancing the adaptability of the model to dynamic working conditions.

[0126] Preprocessing the multi-source monitoring data to obtain standardized data, inputting the standardized data into the dynamic deformation model, optimizing the output of the dynamic deformation model using a deep learning algorithm, establishing a multi-scale prediction mechanism for short-term and long-term deformation trends, and obtaining the deformation prediction results of the hydro-generator set, specifically including:

[0127] Preprocess the multi-source monitoring data to obtain standardized data;

[0128] Remove noise and outliers in the monitoring data through a filtering method to ensure the accuracy and reliability of the data; use the min-max normalization or Z-score normalization method to convert the monitoring data with different dimensions to a unified numerical range, eliminating the influence of dimension differences on subsequent analysis;

[0129] The deep learning algorithm includes a convolutional neural network and a long short-term memory network. The convolutional neural network is used to extract the spatial features in the standardized monitoring data, and the long short-term memory network is used to model the time series dependence of the deformation data, and combine the features extracted by the convolutional neural network to achieve the prediction of multi-scale deformation trends;

[0130] The convolution operation calculation formula of the convolutional neural network is:

[0131]

[0132] Among them, f is the input data, g is the convolution kernel, (i, j) is the output position, and f(m, n) is the input value of the input data at (m, n).

[0133] Specifically, in this embodiment, a filtering method is used to remove noise and outliers in the monitoring data to ensure the accuracy of the input data, and the min-max normalization or Z-score normalization method is used to convert the monitoring data with different dimensions to a unified numerical range, eliminating the influence of dimension differences on the training and prediction of the deep learning model, and providing high-quality input data for feature extraction and time series modeling.

[0134] The convolutional neural network extracts spatial features from the monitoring data through convolutional operations, and can effectively capture the correlations between different physical quantities in multi-source monitoring data. The convolutional kernels of the convolutional neural network can automatically learn the local feature patterns in the monitoring data, such as the coupling relationships between physical quantities like deformation, vibration, and temperature. This enhances the model's ability to extract complex data features, and the convolutional operations ensure the efficiency and accuracy of feature extraction.

[0135] The long short-term memory network captures the time series dependencies of deformation data through its gating mechanism, namely the input gate, forget gate, and output gate. It can model short-term and long-term deformation trends. The long short-term memory network can remember the long-term dependencies of historical data while ignoring irrelevant short-term fluctuations, thereby improving the stability of deformation prediction. Combining the spatial features extracted by the convolutional neural network, the long short-term memory network further enhances the time series modeling ability for deformation data and realizes the prediction of multi-scale deformation trends.

[0136] By combining the advantages of the convolutional neural network and the long short-term memory network, a multi-scale prediction mechanism for short-term and long-term deformation trends is established. Short-term prediction can quickly respond to deformation changes during the operation of the hydro-generator unit, providing support for real-time monitoring and early warning; long-term prediction can provide trend information on deformation, providing a decision-making basis for equipment maintenance and operation optimization. The implementation of the multi-scale prediction mechanism significantly improves the comprehensiveness of deformation prediction.

[0137] Comparing the deformation prediction result with the actually monitored deformation data, calculating the prediction error, adaptively optimizing the model parameters of the dynamic deformation model based on the prediction error, and outputting the optimized deformation prediction result specifically include:

[0138] The weighted absolute error formula is used to calculate the prediction error between the predicted value and the actual value, and the calculation formula is:

[0139]

[0140] where WAE is the weighted absolute error, A is the total number of samples of the monitoring data, ω a is the weight coefficient of the a-th sample, y pred (a) is the predicted deformation value of the a-th sample, and y actual (a) is the actually monitored deformation value of the a-th sample;

[0141] The model parameters of the dynamic deformation model are iteratively updated through the gradient descent method, and the calculation formula is:

[0142]

[0143] where δ T+1 is the model parameter after the (T + 1)-th iteration, δT is the model parameter after the T-th iteration, η is the learning rate, is the gradient of the model parameter δ, λ1 is the regularization coefficient, Δδ T is the adjustment amount of the model parameter in the T-th iteration, μ is the momentum coefficient, Δδ T-1 is the adjustment amount of the model parameter in the (T - 1)-th iteration, is the gradient of the error function with respect to the model parameter δ.

[0144] Specifically, in this embodiment, by comparing the deformation prediction result with the actually monitored deformation data, calculating the prediction error, and evaluating the prediction accuracy of the dynamic deformation model in real time, through the error feedback mechanism, the model parameters are dynamically adjusted, so that the model can gradually correct the prediction deviation and effectively solve the problem of error accumulation caused by fixed parameters in the traditional deformation prediction model.

[0145] Adopting an optimization method based on gradient descent, the model parameters are iteratively adjusted through the error function to ensure that the model parameters can quickly converge to the optimal value. Introducing the regularization coefficient can effectively prevent overfitting of the model parameters and improve the generalization ability of the model. Adding the momentum term makes the update process of the model parameters smoother and avoids the problem of unstable optimization caused by gradient fluctuations, thereby improving the optimization efficiency and stability.

[0146] Controlling the update step size of the model parameters through the learning rate ensures that the optimization process can not only converge quickly but also avoid problems of oscillation or divergence caused by too large a step size. Reasonably setting the learning rate enables the model to quickly adapt to changes in deformation characteristics under complex working conditions and improves the dynamic adaptability of the model.

[0147] By combining error feedback, gradient descent, regularization, and momentum optimization, an efficient error optimization mechanism is constructed. This mechanism can comprehensively consider the current error, historical error, and adjustment amount of the model parameters in each iteration. Through the error optimization mechanism, the dynamic deformation model can continuously improve the prediction performance and adapt to the complex deformation characteristics during the operation of the hydro-generator set.

[0148] Please refer to Figure 2 , the present invention also provides a deformation tracking and prediction system for a hydro-generator set based on model reconstruction, and the system includes:

[0149] A sensor acquisition module, configured to arrange a sensor network at preset key positions of the hydro-generator set, and acquire multi-source monitoring data of the hydro-generator set through the sensor network;

[0150] A static deformation module, configured to construct a static deformation basic model of the hydro-generator set by using the finite element method, and establish a multi-physical field coupling analysis model of the hydro-generator set based on the static deformation basic model and deformation influencing factors;

[0151] A dynamic deformation module, which is used to perform dynamic parameter identification and update on the multi-physical-field coupling analysis model by using the recursive least squares method to obtain a dynamic deformation model;

[0152] A deformation prediction module, which is used to preprocess the multi-source monitoring data to obtain standardized data, input the standardized data into the dynamic deformation model, optimize the output of the dynamic deformation model by using a deep learning algorithm, establish a multi-scale prediction mechanism for short-term and long-term deformation trends, and obtain the deformation prediction result of the hydro-generator unit;

[0153] An error optimization module, which is used to compare the deformation prediction result with the actually monitored deformation data, calculate the prediction error, adaptively optimize the model parameters of the dynamic deformation model based on the prediction error, and output the optimized deformation prediction result.

[0154] Specifically, a deformation tracking and prediction system for a hydro-generator unit based on model reconstruction in this embodiment integrates a multi-source sensor data acquisition module, a data processing module, a dynamic deformation model construction module, a deep learning prediction module, and an error optimization module, collects multi-physical-field data of key parts of the unit in real time, uses multi-physical-field coupling analysis and deep learning algorithms to achieve accurate prediction of deformation, and dynamically optimizes the prediction model through an error feedback mechanism. The overall system has high precision, high real-time performance and adaptive ability, can comprehensively improve the monitoring and prediction level of the operation state of the hydro-generator unit, and provide reliable technical support for the safe operation and maintenance of the equipment.

[0155] The present invention also discloses an electronic device, including: at least one processor, at least one memory communication interface and a bus: wherein, the processor, the memory and the communication interface complete mutual communication through the bus; the memory stores program instructions executable by the processor, and the processor calls the program instructions to implement a deformation tracking and prediction method for a hydro-generator unit based on model reconstruction.

[0156] The present invention also discloses a computer-readable storage medium, the computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to implement all or part of the steps of the deformation tracking and prediction method for a hydro-generator unit based on model reconstruction in the embodiment of the present invention. The storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disc that can store program codes.

[0157] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A deformation tracking and prediction method for hydro-generator units based on model reconstruction, characterized in that It includes the following steps: Arrange a sensor network at preset key positions of the hydro-generator unit, and collect multi-source monitoring data of the hydro-generator unit through the sensor network; Use the finite element method to construct a static deformation basic model of the hydro-generator unit, and establish a multi-physical field coupling analysis model of the hydro-generator unit based on the static deformation basic model and deformation influencing factors; Use the recursive least squares method to identify and update the dynamic parameters of the multi-physical field coupling analysis model to obtain a dynamic deformation model; Preprocess the multi-source monitoring data to obtain standardized data, input the standardized data into the dynamic deformation model, and use a deep learning algorithm to optimize the output of the dynamic deformation model, establish a multi-scale prediction mechanism for short-term and long-term deformation trends, and obtain the deformation prediction result of the hydro-generator unit; Compare the deformation prediction result with the actually monitored deformation data, calculate the prediction error, adaptively optimize the model parameters of the dynamic deformation model based on the prediction error, and output the optimized deformation prediction result.

2. The deformation tracking and prediction method of a hydro-generating unit based on model reconstruction according to claim 1, characterized in that The arranging a sensor network at preset key positions of the hydro-generator unit and collecting multi-source monitoring data of the hydro-generator unit through the sensor network specifically includes: Arrange fiber optic sensors, acceleration sensors and temperature sensors at the upper support of the hydro-generator unit, guide bearing position, key nodes of the rotor and key nodes of the stator to construct a multi-source sensor network; The fiber optic sensors are used to collect deformation data, including radial displacement and axial displacement; the acceleration sensors are used to collect vibration data, including radial vibration and axial vibration; the temperature sensors are used to collect temperature data, including stator temperature and rotor temperature; it also includes pressure sensors for collecting water pressure data; and speed sensors for collecting speed data; Collect multi-source monitoring data in real time through the multi-source sensor network, and perform time series synchronization processing on the multi-source monitoring data; Use the timestamp alignment method to synchronize the multi-source monitoring data, resample the data collected by all sensors at a unified sampling frequency, and supplement the missing data through an interpolation algorithm.

3. The deformation tracking and prediction method of a hydro-generator unit based on model reconstruction according to claim 1, characterized in that The using the finite element method to construct a static deformation basic model of the hydro-generator unit and establishing a multi-physical field coupling analysis model of the hydro-generator unit based on the static deformation basic model and deformation influencing factors specifically includes: Based on the geometric parameters and material parameters of the hydro-generator unit, use the finite element method to construct a static deformation basic model and establish a mapping relationship between the deformation amount and the external load; Use tetrahedral elements to mesh the hydro-generator unit, establish the stiffness equation of the node displacement and the external load, and establish the relationship between the deformation amount and the stress and strain through the strain energy density function; the deformation amount includes radial deformation and axial deformation, and the external load includes gravity load, water pressure load, temperature load and dynamic load; Taking the water pressure load, temperature field distribution and vibration response as boundary conditions, establish a multi-physical field coupling analysis model of the hydro-generator unit; The coupled equations of the water pressure field, temperature field, and vibration field are established. The block matrix method is used to construct the multi-physical field coupling coefficient matrix, and the multi-physical field coupling calculation is carried out through iterative solution.

4. The deformation tracking and prediction method of a hydro-generator set based on model reconstruction according to claim 1, characterized in that The dynamic parameter identification and update of the multi-physical field coupling analysis model are carried out by using the recursive least squares method to obtain a dynamic deformation model, specifically including: Based on multi-source monitoring data, a parameter identification equation is constructed, and the recursive least squares method is used to identify the model parameters of the multi-physical field coupling analysis model in real time to obtain the initial estimated values of the model parameters; The parameter identification equation includes an observation equation and a state equation, and the recursive least squares method uses the weighted least squares criterion; According to the newly obtained monitoring data, the model parameters are recursively updated, and a forgetting factor is introduced to dynamically adjust the weights of historical data to achieve the adaptive optimization of the model parameters; An exponentially decaying forgetting factor is introduced to dynamically adjust the weights of historical data, making the model pay more attention to recent data; by calculating the covariance matrix of parameter estimation, the reliability of parameter identification is evaluated, and the value of the forgetting factor is dynamically adjusted according to the reliability index.

5. The deformation tracking and prediction method for a hydro-generator set based on model reconstruction according to claim 4, wherein The calculation formulas of the observation equation and the state equation are: y(k1) = H(k1)θ(k1) + v(k1); θ(k1) = θ(k1 - 1) + w(k1); where y(k1) is the observation vector at time k1, H(k1) is the observation matrix at time k1, θ(k1) is the parameter vector to be identified at time k1, v(k1) is the observation noise vector at time k1, θ(k1) is the parameter vector at time k1, θ(k1 - 1) is the parameter vector at time k1 - 1, and w(k1) is the parameter evolution noise vector; The calculation formula of the recursive least squares method is: K(k1) = P(k1 - 1)H T (k1)[λ(k1)I + H(k1)P(k1 - 1)H T (k1)] -1 ; Among them, is the parameter vector at the updated time k1, is the parameter vector at time k1 - 1, K(k1) is the gain matrix at time k1, is the residual vector, P(k1 - 1) is the covariance matrix at time k1 - 1, λ(k1) is the forgetting factor at time k1, and I is the identity matrix; The calculation formula of the covariance matrix is: Among them, \(P(k1)\) is the covariance matrix at time \(k1\), \(P(k1 - 1)\) is the covariance matrix at time \(k1-1\), \(\lambda(k1)\) is the forgetting factor at time \(k1\), which is a dynamic forgetting factor, \(\lambda_0\) is the basic forgetting factor, \(\beta_1\) is the adjustment coefficient, \(e(k1)\) is the estimated error vector at time \(k1\), and \(\exp(\cdot)\) is the exponential function; The calculation formula of the reliability index is: where R(k1) is the reliability index at time k1, tr[P(k1)] is the trace of the covariance matrix P(k1), and tr[P(0)] is the trace of the initial covariance matrix.

6. The deformation tracking and prediction method of a hydro-generator unit based on model reconstruction according to claim 1, characterized in that, The multi-source monitoring data is preprocessed to obtain standardized data. The standardized data is input into the dynamic deformation model, and a deep learning algorithm is used to optimize the output of the dynamic deformation model. A multi-scale prediction mechanism for short-term and long-term deformation trends is established to obtain the deformation prediction results of the hydro-generator unit, specifically including: The noise and outliers in the monitoring data are removed by filtering methods, and the minimum-maximum normalization or Z-score normalization method is used to convert the monitoring data with different dimensions to a unified numerical range to obtain standardized data; The deep learning algorithm includes a convolutional neural network and a long short-term memory network. The convolutional neural network is used to extract the spatial features in the standardized monitoring data, and the long short-term memory network is used to model the time series dependence of the deformation data.

7. The deformation tracking and prediction method of a hydro-generator unit based on model reconstruction according to claim 6, wherein The deformation prediction results are compared with the actually monitored deformation data, the prediction error is calculated, the model parameters of the dynamic deformation model are adaptively optimized based on the prediction error, and the optimized deformation prediction results are output, specifically including: The prediction error between the predicted value and the actual value is calculated using the weighted absolute error formula, and the calculation formula is as follows: Among them, WAE is the weighted absolute error, A is the total number of samples of the monitoring data, ω a is the weight coefficient of the a-th sample, y pred (a) is the predicted deformation value of the a-th sample, y actual (a) is the actual monitored deformation value of the a-th sample; The model parameters of the dynamic deformation model are iteratively updated by the gradient descent method, and the calculation formula is as follows: Among them, δ T+1 is the model parameter after the (T + 1)-th iteration, δ T is the model parameter after the T-th iteration, η is the learning rate, is the gradient of the model parameter δ, λ1 is the regularization coefficient, Δδ T is the adjustment amount of the model parameter in the T-th iteration, μ is the momentum coefficient, Δδ T-1 is the adjustment amount of the model parameter in the (T - 1)-th iteration, is the gradient of the error function with respect to the model parameter δ.

8. A deformation tracking and prediction system for a hydro-generating unit based on model reconstruction, characterized in that, The system includes: A sensor acquisition module, which is used to arrange a sensor network at preset key positions of the hydro-generating unit, and collect multi-source monitoring data of the hydro-generating unit through the sensor network; A static deformation module, which is used to construct a static deformation basic model of the hydro-generating unit by using the finite element method, and establish a multi-physical field coupling analysis model of the hydro-generating unit based on the static deformation basic model and deformation influencing factors; A dynamic deformation module, which is used to identify and update the dynamic parameters of the multi-physical field coupling analysis model by using the recursive least squares method to obtain a dynamic deformation model; A deformation prediction module, which is used to preprocess the multi-source monitoring data to obtain standardized data, input the standardized data into the dynamic deformation model, optimize the output of the dynamic deformation model by using a deep learning algorithm, establish a multi-scale prediction mechanism for short-term and long-term deformation trends, and obtain the deformation prediction result of the hydro-generating unit; An error optimization module, which is used to compare the deformation prediction result with the actually monitored deformation data, calculate the prediction error, adaptively optimize the model parameters of the dynamic deformation model based on the prediction error, and output the optimized deformation prediction result.

9. An electronic device, characterized in that, It includes: At least one processor, at least one memory, a communication interface, and a bus; Among them, the processor, the memory, and the communication interface complete communication with each other through the bus. The memory stores program instructions executable by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to implement the method according to any one of claims 1 to 7.

Images