Real-time monitoring method and device for temperature and stress state of supercritical steam turbine rotor

Abstract

The embodiment of the invention provides a supercritical steam turbine rotor temperature and stress state real-time monitoring method and device, and the method comprises the steps: obtaining the operation parameters of a current steam turbine generator unit rotor from a DCS system of a current power plant, and inputting the operation parameters into a pre-trained graph convolution model, the graph convolution model outputs corresponding reconstruction results of a peak-shaving rotor temperature field and a stress field, and according to the reconstruction results of the peak-shaving rotor temperature field and the stress field, the temperature state and the stress state of the current steam turbine generator unit rotor in the peak-shaving operation state are determined; the temperature and stress states of the steam turbine rotor can be monitored in real time under the background of depth peak regulation normalization, and safe and reliable operation of a generator set is guaranteed.

Description

Technical Field

[0001] The present application relates to the field of steam turbine rotor monitoring, and in particular to a real-time monitoring method and device for the temperature and stress state of a supercritical steam turbine rotor. Background Art

[0002] With the increasing investment in new energy generation in the power system, thermal power units have gradually transformed from the main power source to a regulating power source. The power grid requires large-capacity supercritical thermal power units to have the ability to quickly and deeply regulate the peak. When the thermal power unit enters the normalized peak regulation operation, the steam parameters at the turbine inlet will change dramatically. When the steam parameters change dramatically, the temperature distribution inside the turbine rotor is uneven, which causes a sharp increase in the thermal stress level. The long-term action of alternating thermal stress on the rotor will cause the rotor life loss, cracks, bending and fractures, and other faults, which seriously threaten the safety and stability of the unit operation. Therefore, during the normalized deep peak regulation operation of the unit, real-time monitoring of the temperature and stress status of the turbine rotor is of great significance to ensure the safe and reliable operation of the unit and achieve the optimization of the variable load rate. At the same time, it has a positive effect on the life management of the turbine rotor. However, given the harsh working environment of the turbine rotor, it is impossible to use sensors and other methods to monitor the rotor for a long time, so it is necessary to develop indirect monitoring technology.

[0003] In order to overcome the limitation of the lack of real-time performance of the finite element method, two main strategies have been adopted: one adopts a low-precision physical model, and the other chooses to establish a machine learning model to replace the finite element analysis process to quickly predict the rotor physical field and realize real-time monitoring of the temperature and thermal stress of the rotor danger point. For example, polynomial regression (PR), support vector machine (SVM), artificial neural network (ANN), etc. belong to the above methods. However, the accuracy of the physical model is limited by its degree of simplification and the accuracy of the estimation of physical parameters, and the accuracy of the machine learning model is limited by the complexity of the sample data space. The traditional machine learning agent model only focuses on the relationship between the input variables and the prediction target, ignoring the actual physical properties, which is essentially different from the finite element analysis, resulting in significant limitations in the prediction model. Summary of the invention

[0004] In response to the problems in the prior art, the present application provides a real-time monitoring method and device for the temperature and stress state of a supercritical steam turbine rotor, which can monitor the temperature and stress state of the steam turbine rotor in real time under the background of normalized deep peak regulation, and ensure the safe and reliable operation of the generator set.

[0005] In order to solve at least one of the above problems, the present application provides the following technical solutions:

[0006] According to a first aspect of an embodiment of the present application, the present application provides a real-time monitoring method for the temperature and stress state of a supercritical steam turbine rotor, comprising:

[0007] Obtaining the operating parameters of the current turbine generator rotor from the DCS system of the current power plant, wherein the operating parameters include the main steam temperature, extraction temperatures of each stage, exhaust temperature, and the temperature change rates of the main steam and exhaust gas;

[0008] The operating parameters are input into a pre-trained graph convolution model so that the graph convolution model outputs the reconstruction results of the corresponding peak-shaving rotor temperature field and stress field, wherein the graph convolution model is obtained by performing finite element analysis on the peak-shaving operation curve data of the power plant, extracting the data of the rotor temperature and stress state to construct a monitoring database, and performing normalization processing and then training and optimization on the data;

[0009] The temperature and stress states of the current turbine generator rotor in the peak-shaving operation state are determined according to the reconstruction results of the peak-shaving rotor temperature field and stress field, wherein the stress state includes the material yield limit result of the rotor stress danger point.

[0010] According to any embodiment of the present application, the training process of the graph convolution model includes:

[0011] Obtaining the key control points of the actual peak-shaving operation process of the power plant, and generating a plurality of operation curves under different peak-shaving states according to the value range of the key control points;

[0012] Performing finite element analysis on the operation curves under the multiple different peak load regulation states, and constructing a real-time monitoring database for rotor temperature and stress state according to the extracted rotor thermal analysis and strength analysis results;

[0013] Normalizing the data of rotor temperature and stress state in the real-time monitoring database and dividing them into a training set and a validation set;

[0014] According to the training set and the validation set, a temperature field reconstruction model from the main steam temperature, extraction temperatures of each stage, exhaust temperature, main steam temperature change rate, exhaust temperature change rate to the rotor temperature field is constructed, as well as a stress field reconstruction model from the node temperature and node coordinates to the rotor stress field.

[0015] According to any embodiment of the present application, the step of obtaining the key control points of the actual peak load regulation operation process of the power plant and generating a plurality of operation curves under different peak load regulation states according to the value range of the key control points includes:

[0016] According to the power plant operating conditions, determine the key control points of the actual peak load operation process of the power plant;

[0017] Performing linear interpolation on the key control points to obtain a corresponding peak-shaving operation curve of the steam turbine generator set;

[0018] The parameter value range of the key control point is determined based on different peak-shaving requirements, and the peak-shaving operation curve of the steam turbine generator set is sampled based on the Latin hypercube sampling method to obtain the operation curves under the multiple different peak-shaving states.

[0019] According to any embodiment of the present application, the finite element analysis is performed on the operation curves under the multiple different peak-shaving states, and a real-time monitoring database for rotor temperature and stress state is constructed according to the extracted rotor thermal analysis and strength analysis results, including:

[0020] A geometric model is constructed according to the design parameters of the rotor, and based on the geometric model, the centrifugal force of the blade and the annular body and the equivalent radius of the wheel rim are determined by using an equivalent mass block simplification method in which a continuous annular body replaces the blade;

[0021] According to the predefined thermal boundary conditions and mechanical boundary conditions, the finite element method is solved cyclically to obtain the transient stress response results of the temperature field and stress field of the rotor loaded at different time steps;

[0022] Extract corresponding node coordinates, temperature, temperature change rate and stress data according to the response results to build a monitoring database;

[0023] The rotor stress danger point is determined according to the maximum stress position in the monitoring database.

[0024] According to any embodiment of the present application, the data of the rotor temperature and stress state in the real-time monitoring database are normalized and divided into a training set and a validation set, including:

[0025] Normalizing the data of the rotor temperature and stress state based on the maximum and minimum values ​​of the physical quantities in the real-time monitoring database;

[0026] The normalized data sets are randomly sorted, a preset proportion of the data sets are used as the training set, and the remaining proportion of the data sets are used as the validation set.

[0027] According to any embodiment of the present application, the temperature field reconstruction model from the main steam temperature, the extraction temperature of each stage, the exhaust temperature, the main steam temperature change rate, the exhaust temperature change rate to the rotor temperature field and the stress field reconstruction model from the node temperature and the node coordinates to the rotor stress field are constructed according to the training set and the validation set, including:

[0028] A temperature field reconstruction model and a stress field reconstruction model are constructed according to the data of the training set and the validation set, wherein both models adopt the same graph convolutional network structure, including an input layer, an output layer and five intermediate layers, the input layer and the intermediate layer use a Gaussian deviation linear unit activation function, and the output layer does not use an activation function;

[0029] The normalized data set is input into the temperature field reconstruction model and the stress field reconstruction model for training, and the mean square error (MSE) is used as the basic loss function to calculate the difference between the output value of each sample and the actual value, and the reconstruction accuracy of the key stress and temperature points of the model is optimized by the MSE of the field maximum value and the global physical field gradient;

[0030] Among them, the Adamax optimization algorithm is used to set the initial learning rate during the training process, and the learning rate is adaptively adjusted according to the change of the loss function value to improve the training efficiency and stability of the model, so that the corresponding model can reconstruct the data relationship from the main steam temperature, extraction temperatures of each stage, exhaust temperature and its change rate to the rotor temperature field, as well as the data relationship from the node temperature and node coordinates to the rotor stress field.

[0031] According to any embodiment of the present application, the temperature and stress state of the current turbine generator rotor in the peak-shaving operation state is determined according to the reconstruction results of the peak-shaving rotor temperature field and stress field, and the stress state includes the material yield limit result of the rotor stress danger point, including:

[0032] Based on the graph convolution model, the rotor state in peak load operation is analyzed according to the reconstructed temperature field and stress field data;

[0033] Among them, the ratio between the actual stress of the rotor and the design yield limit of the material is used to judge whether the stress of the rotor in the current operating state exceeds the design stress and safety limit, and an operating condition warning is issued when it exceeds the design stress and safety limit.

[0034] According to a second aspect of an embodiment of the present application, the present application provides a real-time monitoring device for temperature and stress state of a supercritical steam turbine rotor, comprising:

[0035] The parameter acquisition module is used to: obtain the operating parameters of the current turbine generator rotor from the DCS system of the current power plant, wherein the operating parameters include the main steam temperature, the extraction temperature of each stage, the exhaust temperature, and the temperature change rate of the main steam and the exhaust gas;

[0036] A model operation module, used to: input the operating parameters into a pre-trained graph convolution model, so that the graph convolution model outputs the reconstruction results of the corresponding peak-shaving rotor temperature field and stress field, wherein the graph convolution model is obtained by performing finite element analysis on the peak-shaving operation curve data of the power plant, extracting the data of the rotor temperature and stress state to construct a monitoring database, and performing normalization processing on the data and then training and optimizing the data;

[0037] The monitoring and early warning module is used to determine the temperature and stress state of the rotor of the current steam turbine generator set under the peak-shaving operation state according to the reconstruction results of the peak-shaving rotor temperature field and stress field, wherein the stress state includes the material yield limit result of the rotor stress danger point.

[0038] According to any embodiment of the present application, the training process of the graph convolution model includes:

[0039] The curve acquisition module is used to: acquire the key control points of the actual peak-shaving operation process of the power plant, and generate a plurality of operation curves under different peak-shaving states according to the value range of the key control points;

[0040] A finite element analysis module, used to: perform finite element analysis on the operation curves under the multiple different peak-shaving states, and build a real-time monitoring database for rotor temperature and stress state according to the extracted rotor thermal analysis and strength analysis results;

[0041] A data normalization module is used to normalize the data of rotor temperature and stress state in the real-time monitoring database and divide them into a training set and a verification set;

[0042] The model training module is used to construct a temperature field reconstruction model from the main steam temperature, extraction temperatures of each stage, exhaust temperature, main steam temperature change rate, exhaust temperature change rate to the rotor temperature field, and a stress field reconstruction model from the node temperature and node coordinates to the rotor stress field according to the training set and the verification set.

[0043] According to any implementation of the present application, the curve acquisition module includes:

[0044] The key control point determination unit is used to: determine the key control points of the actual peak load regulation operation process of the power plant according to the power plant operating conditions;

[0045] A linear interpolation unit is used to: perform linear interpolation on the key control points to obtain a corresponding peak-shaving operation curve of the steam turbine generator set;

[0046] The sampling unit is used to determine the parameter value range of the key control point based on different peak-shaving requirements, and sample the peak-shaving operation curve of the steam turbine generator set based on the Latin hypercube sampling method to obtain the operation curves under the multiple different peak-shaving states.

[0047] According to any implementation of the present application, the finite element analysis module includes:

[0048] A geometric model building unit is used to: build a geometric model according to the design parameters of the rotor, and determine the centrifugal force of the blade and the annular body and the equivalent radius of the rim based on the geometric model by using an equivalent mass block simplification method in which a continuous annular body replaces the blade;

[0049] The finite element solution unit is used to: perform finite element cyclic solution according to predefined thermal boundary conditions and mechanical boundary conditions to obtain transient stress response results of temperature field and stress field of the rotor loaded at different time steps;

[0050] A data extraction unit, used to extract corresponding node coordinates, temperature, temperature change rate and stress data according to the response results to construct a monitoring database;

[0051] The danger point determination unit is used to determine the rotor stress danger point according to the maximum stress position in the monitoring database.

[0052] According to any implementation of the present application, the data normalization module includes:

[0053] A normalization processing unit, used to normalize the data of the rotor temperature and stress state based on the maximum and minimum values ​​of the physical quantities in the real-time monitoring database;

[0054] The data partitioning unit is used to randomly sort the normalized data set, use a preset proportion of the data set as the training set, and use the remaining proportion of the data set as the verification set.

[0055] According to any implementation of the present application, the model training module includes:

[0056] A model construction unit, used to: construct a temperature field reconstruction model and a stress field reconstruction model according to the data of the training set and the validation set, wherein both models adopt the same graph convolutional network structure, including an input layer, an output layer and five intermediate layers, the input layer and the intermediate layer use a Gaussian deviation linear unit activation function, and the output layer does not use an activation function;

[0057] A model training unit is used to: input the normalized data set into the temperature field reconstruction model and the stress field reconstruction model for training, and use the mean square error (MSE) as a basic loss function to calculate the difference between each sample output value and the actual value, and optimize the reconstruction accuracy of the model for key stress and temperature points through the MSE of the field maximum value and the global physical field gradient;

[0058] Among them, the Adamax optimization algorithm is used to set the initial learning rate during the training process, and the learning rate is adaptively adjusted according to the change of the loss function value to improve the training efficiency and stability of the model, so that the corresponding model can reconstruct the data relationship from the main steam temperature, extraction temperatures of each stage, exhaust temperature and its change rate to the rotor temperature field, as well as the data relationship from the node temperature and node coordinates to the rotor stress field.

[0059] According to any implementation of the present application, the monitoring and early warning module includes:

[0060] A rotor state analysis unit, used to: analyze the rotor state in peak load operation according to the reconstructed temperature field and stress field data based on the graph convolution model;

[0061] Among them, the ratio between the actual stress of the rotor and the design yield limit of the material is used to judge whether the stress of the rotor in the current operating state exceeds the design stress and safety limit, and an operating condition warning is issued when it exceeds the design stress and safety limit.

[0062] According to the third aspect of the embodiment of the present application, the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the method for real-time monitoring of the temperature and stress state of a supercritical steam turbine rotor are implemented.

[0063] According to a fourth aspect of an embodiment of the present application, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for real-time monitoring of the temperature and stress state of a supercritical steam turbine rotor.

[0064] According to the fifth aspect of the embodiment of the present application, the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the real-time monitoring method of the temperature and stress state of a supercritical steam turbine rotor.

[0065] It can be seen from the above technical scheme that the present application provides a real-time monitoring method and device for the temperature and stress state of a supercritical steam turbine rotor, by obtaining the operating parameters of the current steam turbine generator set rotor from the DCS system of the current power plant, and inputting the operating parameters into a pre-trained graph convolution model so that the graph convolution model outputs the corresponding reconstruction results of the peak-shaving rotor temperature field and stress field, and determines the temperature and stress state of the current steam turbine generator set rotor under the peak-shaving operating state according to the reconstruction results of the peak-shaving rotor temperature field and stress field; it can monitor the temperature and stress state of the steam turbine rotor in real time under the background of normalized deep peak-shaving, and ensure the safe and reliable operation of the generator set. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0067] Figure 1 This is one of the flow charts of the real-time monitoring method in the embodiment of the present application;

[0068] Figure 2 This is the second flow chart of the real-time monitoring method in the embodiment of the present application;

[0069] Figure 3 This is the third flow chart of the real-time monitoring method in the embodiment of the present application;

[0070] Figure 4 This is a fourth flow chart of the real-time monitoring method in an embodiment of the present application;

[0071] Figure 5 This is the fifth flow chart of the real-time monitoring method in the embodiment of the present application;

[0072] Figure 6 This is the sixth flow chart of the real-time monitoring method in the embodiment of the present application;

[0073] Figure 7 This is the seventh flow chart of the real-time monitoring method in the embodiment of the present application;

[0074] Figure 8 Schematic diagram of the overall architecture of the supercritical steam turbine rotor temperature and stress reconstruction model constructed based on the graph convolutional neural network in the embodiment of the present application;

[0075] Fig. 9 This is a parameterized schematic diagram of a peak-shaving operating curve of a supercritical unit in an embodiment of the present application.

[0076] Fig.10 This is a geometric model diagram of a high-pressure rotor of a supercritical unit in an embodiment of the present application;

[0077] Fig.11 This is a schematic diagram of the location of local stress danger points on the high-pressure rotor of a supercritical unit in an embodiment of the present application.

[0078] Fig.12 This is one of the structural diagrams of the real-time monitoring device in the embodiment of the present application;

[0079] Fig.13 This is the second structural diagram of the real-time monitoring device in the embodiment of the present application;

[0080] Fig.14 It is a schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0082] The acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations.

[0083] Taking into account the limitations of current inspection and monitoring technology, the present application provides a real-time monitoring method and device for the temperature and stress state of a supercritical steam turbine rotor, which can monitor the temperature and stress state of the steam turbine rotor in real time under the background of normalized deep peak regulation to ensure the safe and reliable operation of the generator set.

[0084] In order to monitor the temperature and stress state of the steam turbine rotor in real time under the background of deep peak regulation and ensure the safe and reliable operation of the generator set, the present application provides an embodiment of a method for real-time monitoring of the temperature and stress state of a supercritical steam turbine rotor, see Figure 1 The real-time monitoring method of the temperature and stress state of the supercritical steam turbine rotor specifically includes the following contents:

[0085] Step S101: obtaining the operating parameters of the current turbine generator rotor from the DCS system of the current power plant, wherein the operating parameters include main steam temperature, extraction temperatures of each stage, exhaust temperature, and the temperature change rates of main steam and exhaust.

[0086] Among them, in the real-time monitoring process of the supercritical steam turbine rotor, the key operating parameters of the current steam turbine generator set are first obtained from the DCS (distributed control system) of the power plant. These parameters include the main steam temperature, the extraction temperature and exhaust temperature of each level, and the temperature change rate of the main steam and exhaust. These data reflect the operating characteristics of the steam turbine under different operating conditions. By continuously monitoring these parameters, the dynamic changes of the steam turbine in peak-shaving operation can be captured, providing accurate data input for subsequent state reconstruction.

[0087] Step S102: Input the operating parameters into a pre-trained graph convolution model so that the graph convolution model outputs the corresponding reconstruction results of the peak-shaving rotor temperature field and stress field, wherein the graph convolution model performs finite element analysis on the peak-shaving operation curve data of the power plant, extracts the data of the rotor temperature and stress state to construct a monitoring database, and performs normalization processing and then training and optimization on the data.

[0088] The operating parameters obtained from the DCS system are input into the pre-trained graph convolutional neural network (GCN) model to reconstruct the temperature field and stress field of the turbine rotor. The training process of the graph convolution model is completed through a large amount of curve data in the peak-shaving operation of the power plant: first, the finite element analysis (FEA) of the operating curves under different peak-shaving conditions is performed to extract the temperature and stress field data of the rotor under different load conditions, and a real-time monitoring database is constructed; then, the data in the database is normalized to eliminate the differences between physical quantities, and further divided into training sets and validation sets. The model is optimized and trained through these data sets to improve the accuracy of reconstruction and the generalization ability of the model. After the input operating parameters are calculated by the model, the reconstruction results of the current temperature field and stress field of the rotor are output, providing accurate physical state data for real-time monitoring and early warning.

[0089] Step S103: determining the temperature and stress state of the current turbine generator rotor in the peak-shaving operation state according to the reconstruction results of the peak-shaving rotor temperature field and stress field, wherein the stress state includes the material yield limit result of the rotor stress danger point.

[0090] Among them, according to the reconstruction results of the temperature field and stress field output by the graph convolution model, the system analyzes the peak-shaving operation state of the steam turbine rotor in real time and determines its current temperature distribution and stress state. The stress state not only includes the stress field distribution of the entire rotor, but also focuses on identifying the stress danger points on the rotor, that is, the areas where the stress is concentrated and may exceed the yield limit of the material. By comparing the actual stress value with the yield limit value of the material design, combined with the preset safety factor, the judgment is made to ensure the safe operation of the equipment under high load conditions. If the stress exceeds the design limit, the system will issue an operating condition warning in time to prompt the operator to make necessary operational adjustments or interventions, thereby avoiding equipment damage and operating accidents. The implementation of this step provides real-time safety protection for the peak-shaving operation of the steam turbine.

[0091] From the above description, it can be seen that the real-time monitoring method of the temperature and stress state of the supercritical turbine rotor provided in the embodiment of the present application can monitor the temperature and stress state of the turbine rotor in real time under the background of normalized deep peak regulation, thereby ensuring the safe and reliable operation of the generator set.

[0092] In one embodiment of the real-time monitoring method of the temperature and stress state of the supercritical steam turbine rotor of the present application, see Figure 2 , the training process of the graph convolutional model includes:

[0093] Step S01: Acquire the key control points of the actual peak-shaving operation process of the power plant, and generate a plurality of operation curves under different peak-shaving states according to the value range of the key control points.

[0094] Among them, according to the actual peak-shaving operation curve extracted from the DCS system database of the power plant, the operation curve is parameterized to determine the position of the key control point of the curve, and according to the value range of each parameter in the peak-shaving operation, the Latin hypercube sampling method is used to obtain a large number of operation curves under different peak-shaving conditions.

[0095] In an alternative embodiment, see Figure 3 , the acquisition of the key control points of the actual peak-shaving operation process of the power plant, and the generation of multiple operation curves under different peak-shaving states according to the value range of the key control points, including:

[0096] Step S01A: Determine the key control points of the actual peak load regulation operation process of the power plant according to the power plant operating conditions;

[0097] Step S01B: performing linear interpolation on the key control points to obtain a corresponding peak-shaving operation curve of the steam turbine generator set;

[0098] Step S01C: Determine the parameter value range of the key control point based on different peak-shaving requirements, and sample the peak-shaving operation curve of the steam turbine generator set based on the Latin hypercube sampling method to obtain the operation curves under the multiple different peak-shaving states.

[0099] Specifically, a peak-shaving operation curve parameterization model is first established. The model is based on a set of key points. The given values ​​of the key points under different working conditions are different. Linear interpolation is performed on the basis of the key points to obtain a linear and actual peak-shaving operation curve of the steam turbine generator set.

[0100] Afterwards, according to different peak-shaving requirements, the value range of key parameters in the peak-shaving operation process is determined, and the Latin hypercube sampling method is used for sampling to generate a variety of operating curves under different peak-shaving conditions.

[0101] Based on the above method, the Latin hypercube sampling method used in this application can ensure that the samples in the tail layer within the sampling range will also be sampled, and can reduce the correlation between input variables, thereby improving the prediction accuracy and efficiency of the GCN reconstruction model.

[0102] Step S02: performing finite element analysis on the operation curves under the plurality of different peak load regulation states, and constructing a real-time monitoring database for rotor temperature and stress state according to the extracted rotor thermal analysis and strength analysis results.

[0103] Among them, based on a large number of peak-shaving operation curve samples obtained by sampling, finite element analysis was carried out on the rotor temperature field and stress field under different peak-shaving conditions, the simulated rotor thermal analysis and strength analysis results were extracted, and a real-time monitoring database for rotor temperature and stress state was constructed. The peak-shaving rotor danger points were determined based on the finite element results.

[0104] In an alternative embodiment, see Figure 4 , the finite element analysis is performed on the operation curves under the multiple different peak-shaving states, and a real-time monitoring database for rotor temperature and stress state is constructed according to the extracted rotor thermal analysis and strength analysis results, including:

[0105] Step S02A: constructing a geometric model according to the design parameters of the rotor, and determining the centrifugal force of the blades and the annular body and the equivalent radius of the rim based on the geometric model by using an equivalent mass block simplification method in which a continuous annular body replaces the blades;

[0106] Step S02B: performing finite element cyclic solving according to predefined thermal boundary conditions and mechanical boundary conditions to obtain transient stress response results of temperature field and stress field of the rotor loaded at different time steps;

[0107] Step S02C: extracting corresponding node coordinates, temperature, temperature change rate and stress data according to the response results to construct a monitoring database;

[0108] Step S02D: Determine the rotor stress danger point according to the maximum stress position in the monitoring database.

[0109] Specifically, first, the rotor geometry model is constructed according to the rotor engineering drawing. When considering the centrifugal force load, the simplified method of equivalent mass block is adopted, that is, the blades at each level are replaced by a continuous annular body of the same material. The centrifugal force generated by this annular body is equal to the centrifugal force generated by the original blade. The calculation formula of the blade centrifugal force is as follows:

[0110] C d =mR p ω 2 (1)

[0111] Among them C d is the centrifugal force of the blade; m is the mass of the entire blade; R p is the average radius of the blade; ω is the angular velocity of rotation;

[0112] The calculation formula of equivalent circular centrifugal force is as follows:

[0113]

[0114] Where y is the axial thickness of the ring at radius R; ρ is the material density.

[0115] Integrate the entire ring and according to the centrifugal force equivalent principle, the equivalent radius of the impeller outer edge can be obtained as follows:

[0116]

[0117] After that, finite element solution calculation is carried out. The first thing to do is thermal analysis, which requires defining thermal boundary conditions. Then, structural analysis is carried out, which requires setting mechanical boundary conditions and establishing a cyclic solution process. By loading the rotor temperature field of the corresponding time step respectively, the transient stress field of the rotor under the peak-shaving condition can be obtained.

[0118] According to the finite element solution results, the node coordinates C = [x i ,y i ], node temperature T i , nodal stress σ i , i represents the node number, which constitutes the original database for training graph convolutional neural network [X] = {C, T, σ, T pq ,a pq ,T zq ,a zq ,T cq}, T pq is the exhaust steam temperature under a peak load condition, a pq is the exhaust temperature change rate of a peak load condition, T zq is the main steam temperature under a peak load condition, a zq is the main steam temperature change rate of a certain peak load condition, T cq It is the extraction temperature of each level under a peak load condition, and according to the frequency of the location where the maximum equivalent stress occurs, the rotor stress danger point is determined as the key focus position for real-time monitoring.

[0119] Step S03: normalizing the data of rotor temperature and stress state in the real-time monitoring database, and dividing the data into a training set and a validation set.

[0120] Among them, the data in the real-time monitoring database of rotor temperature and stress state are normalized respectively, and after random shuffling, the data are divided into training set and validation set in a certain proportion as the data set of GCN.

[0121] In an alternative embodiment, see Figure 5 , the data of the rotor temperature and stress state in the real-time monitoring database are normalized and divided into a training set and a validation set, including:

[0122] Step S03A: normalizing the data of the rotor temperature and stress state based on the maximum and minimum values ​​of the physical quantities in the real-time monitoring database;

[0123] Step S03B: randomly sort the normalized data sets, use a preset proportion of the data sets as the training set, and use the remaining proportion of the data sets as the validation set.

[0124] Specifically, in the process of physical field reconstruction, the collected data has clear upper and lower bound characteristics, and accurate prediction of the maximum stress value is also the key to the problem. Therefore, the present invention mainly uses linear normalization to process the original data set;

[0125] The linear normalization processing method is:

[0126]

[0127] Where x max ,x min For the data set x i The maximum and minimum values ​​of the corresponding physical quantity, ε = 1 × 10 -6 , is a small amount, the above formula can compress the data to the interval [0,1];

[0128] Afterwards, the normalized dataset Random sorting is performed and 70% of the data set is used as the training set 30% of the dataset is used as a validation set

[0129] Step S04: construct a temperature field reconstruction model from the main steam temperature, extraction temperatures of each stage, exhaust temperature, main steam temperature change rate, exhaust temperature change rate to the rotor temperature field, and a stress field reconstruction model from the node temperature and node coordinates to the rotor stress field based on the training set and the validation set.

[0130] Among them, a graph convolutional neural network proxy model is constructed, including a temperature field reconstruction model T-GCN and a stress field reconstruction model S-GCN. The data set in step 3 is used for training to obtain the T-GCN temperature field reconstruction model from the main steam temperature, extraction temperatures of each level, exhaust temperature, main steam temperature change rate, exhaust temperature change rate to the rotor temperature field, and the S-GCN stress field reconstruction model from the node temperature and node coordinates to the rotor stress field.

[0131] In an alternative embodiment, see Figure 6 , the temperature field reconstruction model from the main steam temperature, extraction temperatures of each level, exhaust temperature, main steam temperature change rate, exhaust temperature change rate to the rotor temperature field, and the stress field reconstruction model from the node temperature and node coordinates to the rotor stress field are constructed according to the training set and the validation set, including:

[0132] Step S04A: constructing a temperature field reconstruction model and a stress field reconstruction model according to the data of the training set and the validation set, wherein both models adopt the same graph convolutional network structure, including an input layer, an output layer and five intermediate layers, the input layer and the intermediate layer use a Gaussian deviation linear unit activation function, and the output layer does not use an activation function;

[0133] Step S04B: Input the normalized data set into the temperature field reconstruction model and the stress field reconstruction model for training, and use the mean square deviation MSE as the basic loss function to calculate the difference between each sample output value and the actual value. At the same time, the reconstruction accuracy of the key stress and temperature points of the model is optimized through the MSE of the field maximum value and the global physical field gradient.

[0134] Among them, the Adamax optimization algorithm is used to set the initial learning rate during the training process, and the learning rate is adaptively adjusted according to the change of the loss function value to improve the training efficiency and stability of the model, so that the corresponding model can reconstruct the data relationship from the main steam temperature, extraction temperatures of each stage, exhaust temperature and its change rate to the rotor temperature field, as well as the data relationship from the node temperature and node coordinates to the rotor stress field.

[0135] Specifically, the reconstruction model of the temperature field and stress field of the supercritical steam turbine rotor is constructed by two GCNs in series, where the temperature field reconstruction model is T-GCN and the stress field reconstruction model is S-GCN; T-GCN and S-GCN have the same structure, consisting of 7 graph convolutional layers, including 1 input layer, 1 output layer and 5 intermediate layers. The input layer and the intermediate layer both use Gaussian deviation linear unit (GELU) as the activation function, and the output layer does not use an activation function;

[0136] This application uses normalized data sets to train T-GCN and S-GCN, and uses the mean square error (MSE) loss function, which is expressed as follows:

[0137]

[0138] Where N is the sample size of the training samples in the dataset, X i are the output value and actual value of the neural network under the i-th sample, M i is the number of nodes of the i-th sample; since the maximum value of the field and the global physical field gradient are focused on in the calculation of structural strength, the MSE of the maximum value of the field and the MSE of the global physical field gradient are additionally added to the loss function of training T-GCN and S-GCN. The total loss function expression is as follows:

[0139] L T =L f +L s+L gx +L gy (6)

[0140] Where L f is the loss function of the global physical field, L s is the loss function of the field maximum, L gx , L gy are the loss functions of the global physical field gradients in the x-direction and y-direction respectively;

[0141] Among them, optimization algorithms such as Adamax can be used to train T-GCN and S-GCN, set the initial learning rate, and use an adaptive learning rate descent algorithm based on the loss function.

[0142] In one embodiment of the real-time monitoring method of the temperature and stress state of a supercritical steam turbine rotor of the present application, the temperature and stress state of the rotor of the current steam turbine generator set under the peak-shaving operation state are determined according to the reconstruction results of the peak-shaving rotor temperature field and stress field, and the stress state includes the material yield limit result of the rotor stress danger point, and may also specifically include the following contents:

[0143] Based on the graph convolution model, the rotor state in peak load operation is analyzed according to the reconstructed temperature field and stress field data;

[0144] Among them, the ratio between the actual stress of the rotor and the design yield limit of the material is used to judge whether the stress of the rotor in the current operating state exceeds the design stress and safety limit, and an operating condition warning is issued when it exceeds the design stress and safety limit.

[0145] Specifically, the physical field reconstruction model of the T-GCN and S-GCN series trained in the previous steps is used to monitor the temperature and stress state of the supercritical peak-shaving steam turbine unit rotor in real time. Based on the real-time temperature and stress data of the rotor, the peak-shaving rotor dangerous points are monitored in detail. In order to ensure the safety of the overall and local dangerous points of the rotor during the peak-shaving operation of the unit, multiple actual allowable stresses are determined for different parts of the rotor as safety monitoring limits. The specific formula is as follows:

[0146]

[0147] Where σ represents the stress at the dangerous point of the rotor during actual operation (σ i , i represents the dangerous point number) or the maximum stress value (σ max )(MPa), [σ] is the allowable stress of the rotor design (MPa), σ lim is the material yield limit (MPa), s is the safety factor, and the safety factor is determined according to the location and operating conditions.

[0148] Among them, according to the above-mentioned judgment principles, focus on the dangerous point position of the peak-shaving unit rotor and the position of the maximum stress value in the reconstructed stress field to determine whether the stress limit is exceeded under the peak-shaving condition, and at the same time issue a dangerous condition warning.

[0149] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0150] The GCN-based real-time monitoring method for the temperature and stress state of a supercritical steam turbine rotor provided by the present invention adopts a series neural network of T-GCN and S-GCN to construct a high-precision and high-efficiency physical field reconstruction model of a supercritical steam turbine rotor, which can accurately predict the temperature field and stress field distribution of a supercritical steam turbine rotor under peak load conditions in real time. Compared with traditional physical methods and machine learning methods, the present invention has the advantages of stronger adaptability, higher prediction accuracy and faster response speed, and the present invention has obvious advantages in processing high-dimensional complex data.

[0151] Furthermore, the present invention parameterizes the peak-shaving process curve extracted from the DCS control system and uses the Latin hypercube sampling method for sampling, which can obtain a large number of operation curve samples of different peak-shaving processes, and can ensure that the samples in the tail layer within the sampling range can be covered during the sampling process. At the same time, the correlation between input variables can be reduced, thereby improving the prediction accuracy and efficiency of the GCN physical field reconstruction model;

[0152] Furthermore, the present invention constructs a large amount of temperature and stress finite element solution data of peak load conditions into multi-dimensional data, realizes data fusion, and performs pre-processing when constructing the T-GCN and S-GCN neural network structures, which is conducive to the rapid and effective learning of neural networks and the extraction and utilization of data features;

[0153] Furthermore, the real-time monitoring method of the temperature and stress state of the supercritical steam turbine rotor provided by the present invention is implemented based on GCN, and has the ability to process graph structured data that CNN, GNN and traditional machine learning do not have. The prediction network is applicable to any unstructured grid and is applied to the physical field reconstruction of the supercritical steam turbine rotor under peak load conditions.

[0154] Furthermore, the present invention proposes a method for monitoring and distinguishing the key parts of the rotor structure and the location where the maximum stress occurs based on the GCN physical field reconstruction model. The corresponding stress threshold is determined according to the rotor metal material parameters and the working conditions of the supercritical steam turbine rotor. During the peak-shaving operation, a danger warning can be issued based on the rotor stress condition. At the same time, according to the rotor temperature and stress condition, guidance can be provided for adjusting the peak-shaving operation parameters.

[0155] In summary, the present invention provides a real-time monitoring method for the temperature and stress state of a supercritical steam turbine rotor based on GCN, which has high accuracy, fast response, good adaptability, important engineering significance and broad application prospects.

[0156] In order to further illustrate the present solution, the present application also provides a specific application example of a method for real-time monitoring of the temperature and stress state of a supercritical steam turbine rotor using the above-mentioned real-time monitoring device for the temperature and stress state of a supercritical steam turbine rotor, which specifically includes the following contents:

[0157] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments, but this should not be understood as the scope of the above subject matter of the present invention being limited to the following content. Without departing from the above ideas of the present invention, the method of the present invention is not only applicable to the real-time monitoring of the temperature and state of a supercritical steam turbine rotor, but can also be extended to the physical field monitoring of steam turbine rotors of various parameters according to actual conditions.

[0158] Reference Figure 7 The present invention provides a method for real-time monitoring of the temperature and stress state of a supercritical steam turbine rotor based on GCN, comprising the following steps:

[0159] 1. According to the actual peak-shaving operation curve extracted from the DCS system database of the power plant, the operation curve is parameterized to determine the location of the key control points of the curve. According to the value range of each parameter in the peak-shaving operation, the Latin hypercube sampling method is used to obtain a large number of operation curves under different peak-shaving conditions.

[0160] In order to construct a parameterized model of the peak-shaving operation curve, a set of key points is taken as the benchmark, and the specific values ​​of the key points are adjusted according to different operating conditions. On this basis, the linear interpolation method is used to obtain the peak-shaving operation curve of the steam turbine generator set; according to different peak-shaving requirements, the value range of key parameters in the peak-shaving operation process is determined; in order to effectively simulate and cover the unit operating status under different peak-shaving conditions, the Latin hypercube sampling method is used for sampling to generate a series of unit operating curves representing different peak-shaving conditions.

[0161] By parameterizing the peak-shaving operation curve, the present invention can achieve a large number of sampling of the peak-shaving operation curve, so as to obtain the temperature field and stress field of the turbine rotor under a large number of different peak-shaving conditions, and construct a training data set for the GCN graph convolutional neural network, which effectively makes up for the shortage of the actual number of peak-shaving operation conditions and significantly improves the adaptability and prediction accuracy of the GCN reconstruction model.

[0162] 2. Based on a large number of peak-shaving operation curve samples obtained through sampling, finite element analysis is performed on the rotor temperature field and stress field under different peak-shaving conditions, the simulated rotor thermal analysis and strength analysis results are extracted, a real-time monitoring database for rotor temperature and stress status is constructed, and the peak-shaving rotor danger points are determined based on the finite element results.

[0163] The finite element thermal analysis first conducts steady-state thermal analysis, which requires the definition of thermal boundary conditions that do not change with time. The steam flow area on the outer surface of the rotor is described by the third-class boundary conditions, and the surface heat transfer coefficient is calculated using empirical formulas (Westinghouse empirical formula, Alstom empirical formula). The rotor bearings are regarded as first-class boundary conditions, and the shaft ends and the axis are regarded as adiabatic boundary conditions. After the boundary conditions are set, the calculation is solved to obtain the steady-state temperature field of the rotor; then the transient thermal analysis is carried out, and the rotor temperature obtained from the steady-state thermal analysis is used as the initial temperature boundary condition of the transient thermal analysis. The rotor surface convection heat transfer coefficient and steam temperature that change with time are loaded to solve the rotor transient temperature field;

[0164] After obtaining the transient temperature field of the rotor, the structural finite element analysis is performed on the rotor model. The mechanical boundary conditions are set according to the actual situation of the unit, and a cyclic solution process is established. The rotor temperature field of the corresponding time step is loaded respectively, and the transient stress field of the rotor under the peak-shaving condition can be obtained.

[0165] 3. The data in the real-time monitoring database of rotor temperature and stress state are normalized respectively, and after random shuffling, the data are divided into training set and validation set according to a certain ratio as the data set of GCN.

[0166] In the reconstruction of the rotor physical field, the data set used has parameters with different attributes and different value ranges, and their distribution has large differences, which will affect the convergence speed of the GCN training process. Therefore, the data needs to be normalized before training to reduce the distribution differences of the data and eliminate the dimensional differences between different physical quantities. In the process of physical field reconstruction, the collected data has clear upper and lower bounds, and accurate prediction of the maximum stress value is also the key to the problem. Therefore, the present invention mainly uses linear normalization to process the original data set.

[0167] 4. Construct a graph convolutional neural network proxy model, including a temperature field reconstruction model T-GCN and a stress field reconstruction model S-GCN. Use the data set in step 3 for training to obtain the T-GCN temperature field reconstruction model from the main steam temperature, extraction temperatures at each level, exhaust temperature, main steam temperature change rate, exhaust temperature change rate to the rotor temperature field, and the S-GCN stress field reconstruction model from the node temperature and node coordinates to the rotor stress field.

[0168] During the peak load regulation process, the reconstruction model of the temperature field and stress field of the supercritical steam turbine rotor is constructed by two GCNs in series, where the temperature field reconstruction model is T-GCN and the stress field reconstruction model is S-GCN. Specifically, both T-GCN and S-GCN use the GMM graph convolution model, which is a spatial domain graph convolution that introduces a Gaussian kernel function into the convolution kernel. The convolution operation process is described as follows:

[0169]

[0170] In the formula is the weight parameter of the spatial convolution kernel, m q is the number of neighbor nodes of the qth node.

[0171] The overall architecture of the physical field reconstruction model of T-GCN and S-GCN in series can be referenced Figure 8 :

[0172] T-GCN and S-GCN have the same structure, consisting of 7 graph convolutional layers, including 1 input layer, 1 output layer, and 5 intermediate layers. The input layer and the intermediate layer both use Gaussian deviation linear unit (GELU) as the activation function, and the output layer does not use an activation function. The expression of the GELU activation function is as follows:

[0173] f(x)=xΦ(x)

[0174]

[0175] Where Φ(x) refers to the cumulative function of the Gaussian normal distribution of x; some researchers have given its approximate expression as follows:

[0176]

[0177] The input size of T-GCN is 4+n (n represents the number of extraction stages), which includes the main steam temperature, extraction temperatures of each stage, exhaust temperature, main steam temperature change rate and exhaust temperature change rate. The output size is 1, which is the temperature of the rotor. The input size of S-GCN is 3, which includes the x and y coordinates of the nodes and the node temperature. The output size is 3, which is the three stress components of the rotor.

[0178] The normalized dataset is used to train T-GCN and S-GCN, and both T-GCN and S-GCN use the mean squared error (MSE) loss function.

[0179] Adamax and other optimization algorithms are used to train T-GCN and S-GCN, set the initial learning rate, and use an adaptive learning rate descent algorithm based on the loss function.

[0180] The present invention uses a rotor temperature field and stress field reconstruction model of peak-shaving conditions using two GCNs connected in series, which has the advantages of fast response and high precision. It can realize high-precision and high-efficiency real-time monitoring of the temperature and stress conditions of the turbine generator rotor under peak-shaving operation, laying a foundation for the safe operation of the peak-shaving unit and the intelligent control of the peak-shaving unit.

[0181] 5. Use the GCN-based reconstruction model to reconstruct the temperature field and stress field of the rotor under peak-shaving operation. Based on the reconstruction results, monitor the temperature and stress conditions of the turbine generator rotor in real time under peak-shaving operation. Focus on monitoring the peak-shaving rotor danger points and maximum stress positions determined in step 2, and perform material yield limit warning to ensure the safety of the rotor during peak-shaving operation.

[0182] Using the physical field reconstruction model of T-GCN and S-GCN in series trained in step 4, the temperature and stress state of the supercritical peak-shaving steam turbine unit rotor is monitored in real time. The global temperature field and stress field of the rotor are reconstructed in real time through the steam parameters of the unit's peak-shaving operation, and the temperature and stress data are extracted for storage.

[0183] According to the real-time data of rotor temperature and stress, the peak-shaving rotor dangerous points determined by finite element analysis in step 2 are monitored. In order to ensure the safety of the overall and local dangerous points of the rotor during the peak-shaving operation of the unit, multiple actual allowable stresses are determined for different parts of the rotor as safety monitoring limits.

[0184] According to the established judgment principles, we focus on the potential dangerous points of the peak-shaving unit rotor and the position of the maximum stress in the reconstructed stress field to determine whether there is a risk of stress exceeding the limit value. At the same time, we issue early warnings for dangerous operating conditions to ensure the safety and stability of the unit operation.

[0185] During the peak-shaving operation of the unit, the present invention can monitor the stress state of the rotor in real time. If the stress level approaches or exceeds a preset threshold, a danger warning mechanism is immediately triggered. In addition, the present invention can also provide guidance for adjusting the peak-shaving operation parameters according to the real-time temperature and stress conditions of the rotor, thereby achieving safe and efficient operation of the unit.

[0186] Exemplarily, the real-time monitoring method for temperature and stress state of a supercritical steam turbine rotor based on GCN of the present invention is used to monitor the temperature and stress state of a high-pressure rotor of a supercritical steam turbine unit in real time, as follows:

[0187] 1. If Fig. 9 As shown, according to the method of the present invention, firstly, parameterized modeling is performed on the exhaust steam temperature curve during the peak regulation process, which is specifically described as follows:

[0188] T=a i (tt i )+Ti,out ,t i ≤t<t i+1 ,i=1,2,...,N s

[0189] Where T is exhaust temperature, t is time, N is s is the number of segments; a i With T i,out are the temperature change rate within the i-th period and the initial exhaust temperature of the stage, respectively; therefore, the exhaust temperature T is a piecewise function of time t, and the exhaust temperature and time are in a linear relationship within each period.

[0190] Then, Latin hypercube sampling is used to complete sampling of different working conditions according to the range of parameter values. The sampling process needs to satisfy the following relationship:

[0191]

[0192] Where T0 is the initial exhaust temperature and Δt is the corresponding temperature change time.

[0193] 2. First, model the turbine high-pressure rotor according to the engineering drawing. The specific model is as follows: Fig.10 As shown; then based on the peak-shaving curve sample, the finite element solution under different peak-shaving conditions is performed, and the results include the stress σ in the three directions of each node x , σ y , σ z , the temperature of each node T, these result data are combined with the working condition parameters and node coordinates, and the data under multiple working conditions are fused to obtain the original database [X] = {C, T, σ, T pq ,a pq ,T zq ,a zq ,T cq}, where C is the node coordinate, T is the node temperature, σ is the node stress, T pq is the exhaust steam temperature under a peak load condition, a pq is the exhaust temperature change rate of a peak load condition, T zq is the main steam temperature under a peak load condition, a zq is the main steam temperature change rate of a certain peak load condition, T cq are the first and second stage extraction steam temperatures under a peak load condition. In addition, based on the finite element solution results, the position of the stress danger point of the rotor under the peak load condition is determined. The specific position is as follows: Fig.11 shown.

[0194] 3. Normalize the data in the original database and compress the data to the interval [0,1] to reduce the distribution difference of the data and eliminate the dimensional difference between different physical quantities; then use the normalized data set Random sorting is performed and 70% of the data set is used as the training set 30% of the dataset is used as a validation set

[0195] 4. Construct a GCN-based supercritical steam turbine rotor physical field reconstruction model, train and verify the neural network through the data set, and obtain a high-precision reconstruction model. First, the steam parameters of each peak-shaving condition of the unit are used as the input of T-GCN to reconstruct the temperature field of the rotor; then the reconstructed temperature field is used as the input of S-GCN, and all node temperatures use the data format of the graph to reconstruct the stress field of the rotor.

[0196] Specifically, the input size of T-GCN is 6, which includes the main steam temperature, the first and second stage extraction temperatures, the high pressure cylinder exhaust temperature, the main steam temperature change rate, and the high pressure cylinder exhaust temperature change rate. The output size is 1, which is the temperature of the rotor. The input size of S-GCN is 3, which includes the x and y coordinates of the nodes and the node temperature. The output size is 3, which is the three stress components of the rotor. When training the neural network, the initial learning rate is set to 0.005. Both T-GCN and S-GCN use the Adamax optimization algorithm, and the loss function uses the MSE loss function. The batch sample size is 64, and the adaptive learning rate descent algorithm based on the loss function is used.

[0197] 5. First, based on the steam parameter information of the unit's current operation, the physical field reconstruction model is used to reconstruct the temperature field and stress field of the steam turbine rotor under peak operation; then, based on the temperature and stress reconstruction results, the temperature and stress state of the steam turbine rotor are monitored in real time to determine whether the stress at the rotor's dangerous point and the maximum stress value exceeds the allowable limit, and at the same time, an early warning of dangerous conditions is issued to ensure the safety and stability of the unit's operation. The judgment principles are as follows:

[0198]

[0199] Where σ represents the stress at the dangerous point of the rotor during actual operation (σi, i represents the number of the dangerous point) or the maximum stress (σ max )(MPa), [σ] is the allowable stress of the rotor design (MPa), σ lim is the maximum allowable stress for rotor deformation and damage (MPa), s is the safety factor, and the safety factor is determined according to the location and operating conditions.

[0200] In order to monitor the temperature and stress state of the steam turbine rotor in real time under the background of deep peak regulation and ensure the safe and reliable operation of the generator set, the present application provides an embodiment of a real-time monitoring device for the temperature and stress state of a supercritical steam turbine rotor for realizing all or part of the content of the real-time monitoring method for the temperature and stress state of a supercritical steam turbine rotor, see Fig.12 The real-time monitoring device for the temperature and stress state of the supercritical steam turbine rotor specifically includes the following contents:

[0201] The parameter acquisition module 1101 is used to: acquire the operating parameters of the current turbine generator rotor from the DCS system of the current power plant, wherein the operating parameters include the main steam temperature, the extraction temperatures of each stage, the exhaust temperature, and the temperature change rates of the main steam and the exhaust gas;

[0202] The model operation module 1102 is used to: input the operating parameters into a pre-trained graph convolution model so that the graph convolution model outputs the reconstruction results of the corresponding peak-shaving rotor temperature field and stress field, wherein the graph convolution model is obtained by performing finite element analysis on the peak-shaving operation curve data of the power plant, extracting the data of the rotor temperature and stress state to construct a monitoring database, and performing normalization processing on the data and then training and optimizing the data;

[0203] The monitoring and early warning module 1103 is used to determine the temperature and stress state of the rotor of the current steam turbine generator set under the peak-shaving operation state according to the reconstruction results of the peak-shaving rotor temperature field and stress field, wherein the stress state includes the material yield limit result of the rotor stress danger point.

[0204] According to any embodiment of the present application, see Fig.13 , the training process of the graph convolutional model includes:

[0205] The curve acquisition module 0001 is used to: acquire the key control points of the actual peak-shaving operation process of the power plant, and generate a plurality of operation curves under different peak-shaving states according to the value range of the key control points;

[0206] The finite element analysis module 0002 is used to: perform finite element analysis on the operation curves under the multiple different peak load regulation states, and build a real-time monitoring database for rotor temperature and stress state according to the extracted rotor thermal analysis and strength analysis results;

[0207] The data normalization module 0003 is used to normalize the data of the rotor temperature and stress state in the real-time monitoring database and divide them into a training set and a verification set;

[0208] Model training module 0004 is used to: construct a temperature field reconstruction model from the main steam temperature, extraction temperatures of each stage, exhaust temperature, main steam temperature change rate, exhaust temperature change rate to the rotor temperature field according to the training set and the verification set, and a stress field reconstruction model from the node temperature and node coordinates to the rotor stress field.

[0209] According to any implementation of the present application, the curve acquisition module includes:

[0210] The key control point determination unit is used to: determine the key control points of the actual peak load regulation operation process of the power plant according to the power plant operating conditions;

[0211] A linear interpolation unit is used to: perform linear interpolation on the key control points to obtain a corresponding peak-shaving operation curve of the steam turbine generator set;

[0212] The sampling unit is used to determine the parameter value range of the key control point based on different peak-shaving requirements, and sample the peak-shaving operation curve of the steam turbine generator set based on the Latin hypercube sampling method to obtain the operation curves under the multiple different peak-shaving states.

[0213] According to any implementation of the present application, the finite element analysis module includes:

[0214] A geometric model building unit is used to: build a geometric model according to the design parameters of the rotor, and determine the centrifugal force of the blade and the annular body and the equivalent radius of the rim based on the geometric model by using an equivalent mass block simplification method in which a continuous annular body replaces the blade;

[0215] The finite element solution unit is used to: perform finite element cyclic solution according to predefined thermal boundary conditions and mechanical boundary conditions to obtain transient stress response results of temperature field and stress field of the rotor loaded at different time steps;

[0216] A data extraction unit, used to extract corresponding node coordinates, temperature, temperature change rate and stress data according to the response results to construct a monitoring database;

[0217] The danger point determination unit is used to determine the rotor stress danger point according to the maximum stress position in the monitoring database.

[0218] According to any implementation of the present application, the data normalization module includes:

[0219] A normalization processing unit, used to normalize the data of the rotor temperature and stress state based on the maximum and minimum values ​​of the physical quantities in the real-time monitoring database;

[0220] The data partitioning unit is used to randomly sort the normalized data set, use a preset proportion of the data set as the training set, and use the remaining proportion of the data set as the verification set.

[0221] According to any implementation of the present application, the model training module includes:

[0222] A model construction unit, used to: construct a temperature field reconstruction model and a stress field reconstruction model according to the data of the training set and the validation set, wherein both models adopt the same graph convolutional network structure, including an input layer, an output layer and five intermediate layers, the input layer and the intermediate layer use a Gaussian deviation linear unit activation function, and the output layer does not use an activation function;

[0223] A model training unit is used to: input the normalized data set into the temperature field reconstruction model and the stress field reconstruction model for training, and use the mean square error (MSE) as a basic loss function to calculate the difference between each sample output value and the actual value, and optimize the reconstruction accuracy of the model for key stress and temperature points through the MSE of the field maximum value and the global physical field gradient;

[0224] Among them, the Adamax optimization algorithm is used to set the initial learning rate during the training process, and the learning rate is adaptively adjusted according to the change of the loss function value to improve the training efficiency and stability of the model, so that the corresponding model can reconstruct the data relationship from the main steam temperature, extraction temperatures of each stage, exhaust temperature and its change rate to the rotor temperature field, as well as the data relationship from the node temperature and node coordinates to the rotor stress field.

[0225] According to any implementation of the present application, the monitoring and early warning module includes:

[0226] A rotor state analysis unit, used to: analyze the rotor state in peak load operation according to the reconstructed temperature field and stress field data based on the graph convolution model;

[0227] Among them, the ratio between the actual stress of the rotor and the design yield limit of the material is used to judge whether the stress of the rotor in the current operating state exceeds the design stress and safety limit, and an operating condition warning is issued when it exceeds the design stress and safety limit.

[0228] From the above description, it can be seen that the real-time monitoring device for the temperature and stress state of the supercritical turbine rotor provided in the embodiment of the present application can monitor the temperature and stress state of the turbine rotor in real time under the background of normalized deep peak regulation, thereby ensuring the safe and reliable operation of the generator set.

[0229] Fig.14 A schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention, such as Fig.14 As shown, the electronic device includes: a processor (processor) 301, a memory (memory) 302 and a bus 303.

[0230] The processor 301 and the memory 302 communicate with each other via the bus 303 .

[0231] The processor 301 is used to call the program instructions in the memory 302 to execute the methods provided by the above method embodiments.

[0232] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the real-time monitoring method of the temperature and stress state of the supercritical steam turbine rotor is implemented.

[0233] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the real-time monitoring method of the temperature and stress state of the supercritical steam turbine rotor is implemented.

[0234] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A real-time monitoring method for the temperature and stress state of a supercritical steam turbine rotor, characterized in that: The method comprises: Obtaining the operating parameters of the current turbine generator rotor from the DCS system of the current power plant, wherein the operating parameters include the main steam temperature, extraction temperatures of each stage, exhaust temperature, and the temperature change rates of the main steam and exhaust gas; The operating parameters are input into a pre-trained graph convolution model so that the graph convolution model outputs the reconstruction results of the corresponding peak-shaving rotor temperature field and stress field, wherein the graph convolution model is obtained by performing finite element analysis on the peak-shaving operation curve data of the power plant, extracting the data of the rotor temperature and stress state to construct a monitoring database, and performing normalization processing and then training and optimization on the data; The temperature and stress states of the current turbine generator rotor in the peak-shaving operation state are determined according to the reconstruction results of the peak-shaving rotor temperature field and stress field, wherein the stress state includes the material yield limit result of the rotor stress danger point.

2. The real-time monitoring method for the temperature and stress state of a supercritical steam turbine rotor according to claim 1, characterized in that: The training process of the graph convolutional model includes: Obtaining the key control points of the actual peak-shaving operation process of the power plant, and generating a plurality of operation curves under different peak-shaving states according to the value range of the key control points; Performing finite element analysis on the operation curves under the multiple different peak load regulation states, and constructing a real-time monitoring database for rotor temperature and stress state according to the extracted rotor thermal analysis and strength analysis results; Normalizing the data of rotor temperature and stress state in the real-time monitoring database and dividing them into a training set and a validation set; According to the training set and the validation set, a temperature field reconstruction model from the main steam temperature, extraction temperatures of each stage, exhaust temperature, main steam temperature change rate, exhaust temperature change rate to the rotor temperature field is constructed, as well as a stress field reconstruction model from the node temperature and node coordinates to the rotor stress field.

3. The real-time monitoring method for the temperature and stress state of a supercritical steam turbine rotor according to claim 2, characterized in that: The step of obtaining the key control points of the actual peak load regulation operation process of the power plant and generating a plurality of operation curves under different peak load regulation states according to the value range of the key control points includes: According to the power plant operating conditions, determine the key control points of the actual peak load operation process of the power plant; Performing linear interpolation on the key control points to obtain a corresponding peak-shaving operation curve of the steam turbine generator set; The parameter value range of the key control point is determined based on different peak-shaving requirements, and the peak-shaving operation curve of the steam turbine generator set is sampled based on the Latin hypercube sampling method to obtain the operation curves under the multiple different peak-shaving states.

4. The real-time monitoring method for the temperature and stress state of a supercritical steam turbine rotor according to claim 2, characterized in that: The finite element analysis is performed on the operation curves under the multiple different peak load regulation states, and a real-time monitoring database for rotor temperature and stress state is constructed according to the extracted rotor thermal analysis and strength analysis results, including: A geometric model is constructed according to the design parameters of the rotor, and based on the geometric model, the centrifugal force of the blade and the annular body and the equivalent radius of the wheel rim are determined by using an equivalent mass block simplification method in which a continuous annular body replaces the blade; Finite element cyclic solution is performed according to the predefined thermal boundary conditions and mechanical boundary conditions to obtain the transient stress response results of the temperature field and stress field of the loaded rotor at different time steps; Extract corresponding node coordinates, temperature, temperature change rate and stress data according to the response results to build a monitoring database; The rotor stress danger point is determined according to the maximum stress position in the monitoring database.

5. The real-time monitoring method for the temperature and stress state of a supercritical steam turbine rotor according to claim 2, characterized in that: The data of the rotor temperature and stress state in the real-time monitoring database are normalized and divided into a training set and a validation set, including: Normalizing the data of the rotor temperature and stress state based on the maximum and minimum values ​​of the physical quantities in the real-time monitoring database; The normalized data sets are randomly sorted, a preset proportion of the data sets are used as the training set, and the remaining proportion of the data sets are used as the validation set.

6. The real-time monitoring method for the temperature and stress state of a supercritical steam turbine rotor according to claim 2, characterized in that: The temperature field reconstruction model from the main steam temperature, extraction temperatures of each level, exhaust temperature, main steam temperature change rate, exhaust temperature change rate to the rotor temperature field and the stress field reconstruction model from the node temperature and node coordinates to the rotor stress field are constructed according to the training set and the validation set, including: A temperature field reconstruction model and a stress field reconstruction model are constructed according to the data of the training set and the validation set, wherein both models adopt the same graph convolutional network structure, including an input layer, an output layer and five intermediate layers, the input layer and the intermediate layer use a Gaussian deviation linear unit activation function, and the output layer does not use an activation function; The normalized data set is input into the temperature field reconstruction model and the stress field reconstruction model for training, and the mean square error (MSE) is used as the basic loss function to calculate the difference between the output value of each sample and the actual value, and the reconstruction accuracy of the key stress and temperature points of the model is optimized by the MSE of the field maximum value and the global physical field gradient; Among them, the Adamax optimization algorithm is used to set the initial learning rate during the training process, and the learning rate is adaptively adjusted according to the change of the loss function value to improve the training efficiency and stability of the model, so that the corresponding model can reconstruct the data relationship from the main steam temperature, extraction temperatures of each stage, exhaust temperature and its change rate to the rotor temperature field, as well as the data relationship from the node temperature and node coordinates to the rotor stress field.

7. The real-time monitoring method for the temperature and stress state of a supercritical steam turbine rotor according to claim 1, characterized in that: The temperature and stress state of the current turbine generator rotor under the peak-shaving operation state is determined according to the reconstruction results of the peak-shaving rotor temperature field and stress field, wherein the stress state includes the material yield limit result of the rotor stress danger point, including: Based on the graph convolution model, the rotor state in peak load operation is analyzed according to the reconstructed temperature field and stress field data; Among them, the ratio between the actual stress of the rotor and the design yield limit of the material is used to judge whether the stress of the rotor in the current operating state exceeds the design stress and safety limit, and an operating condition warning is issued when it exceeds the design stress and safety limit.

8. A real-time monitoring device for the temperature and stress state of a supercritical steam turbine rotor, characterized in that: The device comprises: The parameter acquisition module is used to: obtain the operating parameters of the current turbine generator rotor from the DCS system of the current power plant, wherein the operating parameters include the main steam temperature, the extraction temperature of each stage, the exhaust temperature, and the temperature change rate of the main steam and the exhaust gas; A model operation module, used to: input the operating parameters into a pre-trained graph convolution model, so that the graph convolution model outputs the reconstruction results of the corresponding peak-shaving rotor temperature field and stress field, wherein the graph convolution model is obtained by performing finite element analysis on the peak-shaving operation curve data of the power plant, extracting the data of the rotor temperature and stress state to construct a monitoring database, and performing normalization processing on the data and then training and optimizing the data; The monitoring and early warning module is used to determine the temperature and stress state of the rotor of the current steam turbine generator set under the peak-shaving operation state according to the reconstruction results of the peak-shaving rotor temperature field and stress field, wherein the stress state includes the material yield limit result of the rotor stress danger point.

9. The real-time monitoring device for the temperature and stress state of a supercritical steam turbine rotor according to claim 8, characterized in that: The training process of the graph convolutional model includes: The curve acquisition module is used to: acquire the key control points of the actual peak-shaving operation process of the power plant, and generate a plurality of operation curves under different peak-shaving states according to the value range of the key control points; A finite element analysis module, used to: perform finite element analysis on the operation curves under the multiple different peak-shaving states, and build a real-time monitoring database for rotor temperature and stress state according to the extracted rotor thermal analysis and strength analysis results; A data normalization module is used to normalize the data of rotor temperature and stress state in the real-time monitoring database and divide the data into a training set and a verification set; The model training module is used to construct a temperature field reconstruction model from the main steam temperature, extraction temperatures of each stage, exhaust temperature, main steam temperature change rate, exhaust temperature change rate to the rotor temperature field, and a stress field reconstruction model from the node temperature and node coordinates to the rotor stress field according to the training set and the verification set.

10. The real-time monitoring device for the temperature and stress state of a supercritical steam turbine rotor according to claim 9, characterized in that: The curve acquisition module comprises: The key control point determination unit is used to: determine the key control points of the actual peak load regulation operation process of the power plant according to the power plant operating conditions; A linear interpolation unit is used to: perform linear interpolation on the key control points to obtain a corresponding peak-shaving operation curve of the steam turbine generator set; The sampling unit is used to determine the parameter value range of the key control point based on different peak-shaving requirements, and sample the peak-shaving operation curve of the steam turbine generator set based on the Latin hypercube sampling method to obtain the operation curves under the multiple different peak-shaving states.

11. The real-time monitoring device for the temperature and stress state of a supercritical steam turbine rotor according to claim 9, characterized in that: The finite element analysis module includes: A geometric model building unit is used to: build a geometric model according to the design parameters of the rotor, and determine the centrifugal force of the blade and the annular body and the equivalent radius of the rim based on the geometric model by using an equivalent mass block simplification method in which a continuous annular body replaces the blade; The finite element solution unit is used to: perform finite element cyclic solution according to predefined thermal boundary conditions and mechanical boundary conditions to obtain transient stress response results of temperature field and stress field of the rotor loaded at different time steps; A data extraction unit, used to extract corresponding node coordinates, temperature, temperature change rate and stress data according to the response results to construct a monitoring database; The danger point determination unit is used to determine the rotor stress danger point according to the maximum stress position in the monitoring database.

12. The real-time monitoring device for the temperature and stress state of a supercritical steam turbine rotor according to claim 9, characterized in that: The data normalization module comprises: A normalization processing unit, used to normalize the data of the rotor temperature and stress state based on the maximum and minimum values ​​of the physical quantities in the real-time monitoring database; The data partitioning unit is used to randomly sort the normalized data set, use a preset proportion of the data set as the training set, and use the remaining proportion of the data set as the verification set.

13. The real-time monitoring device for the temperature and stress state of a supercritical steam turbine rotor according to claim 9, characterized in that: The model training module includes: A model construction unit, used to: construct a temperature field reconstruction model and a stress field reconstruction model according to the data of the training set and the validation set, wherein both models adopt the same graph convolutional network structure, including an input layer, an output layer and five intermediate layers, the input layer and the intermediate layer use a Gaussian deviation linear unit activation function, and the output layer does not use an activation function; A model training unit is used to: input the normalized data set into the temperature field reconstruction model and the stress field reconstruction model for training, and use the mean square error (MSE) as a basic loss function to calculate the difference between each sample output value and the actual value, and optimize the reconstruction accuracy of the model for key stress and temperature points through the MSE of the field maximum value and the global physical field gradient; Among them, the Adamax optimization algorithm is used to set the initial learning rate during the training process, and the learning rate is adaptively adjusted according to the change of the loss function value to improve the training efficiency and stability of the model, so that the corresponding model can reconstruct the data relationship from the main steam temperature, extraction temperatures of each stage, exhaust temperature and its change rate to the rotor temperature field, as well as the data relationship from the node temperature and node coordinates to the rotor stress field.

14. The real-time monitoring device for the temperature and stress state of a supercritical steam turbine rotor according to claim 8, characterized in that: The monitoring and early warning module includes: A rotor state analysis unit, used to: analyze the rotor state in peak load operation according to the reconstructed temperature field and stress field data based on the graph convolution model; Among them, the ratio between the actual stress of the rotor and the design yield limit of the material is used to judge whether the stress of the rotor in the current operating state exceeds the design stress and safety limit, and an operating condition warning is issued when it exceeds the design stress and safety limit.

15. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the real-time monitoring method of the temperature and stress state of the supercritical steam turbine rotor as described in any one of claims 1 to 7 are implemented.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the real-time monitoring method of the temperature and stress state of a supercritical steam turbine rotor as described in any one of claims 1 to 7 are implemented.

17. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the real-time monitoring method of the temperature and stress state of a supercritical steam turbine rotor as described in any one of claims 1 to 7 are implemented.

Images