A Health Status Assessment Method for Mixed-Flow Hydropower Units Based on Physical Spatiotemporal Weighted Maps

Abstract

This invention relates to the field of health status assessment of hydro-turbine units, and specifically to a method for assessing the health status of mixed-flow hydro-turbine units based on a physical-spatiotemporal weighted graph. The method involves constructing a geometric model of the mixed-flow hydro-turbine unit and calculating simulated pulsation signals. Based on the spatial coordinates and feature data of the pressure pulsation signals, a feature matrix and an adjacency matrix are constructed to capture physical spatial relationships and basic features. Temporal statistical parameters are used to assign weights to the feature relationships and physical spatial relationships. These relationships are then fused to generate a physical-spatiotemporal weighted graph, which is input into a graph convolution model to obtain the health status model of the mixed-flow hydro-turbine unit. Finally, the Euclidean distance between the predicted label and the actual observed signal under degraded conditions is calculated to derive a degradation status index to assess the health status of the mixed-flow hydro-turbine unit. This invention improves the level of health status assessment for mixed-flow hydro-turbine units by integrating prior knowledge of the mixed-flow hydro-turbine unit and combining it with a weighted graph representation learning method.

Description

Technical Field

[0001] This invention relates to the field of health status assessment of hydro turbine units, and more specifically, to a method for assessing the health status of mixed-flow hydro turbine units based on a physical spatiotemporal weighted graph. Background Technology

[0002] Hydropower generating units, as the core equipment for hydropower energy conversion, undertake critical tasks such as peak shaving, frequency regulation, and emergency backup in the power system. Their operating status directly affects the safety and economic benefits of hydropower plants. Due to the frequent switching of operating conditions and the harsh operating environment of hydropower generating units, and the interdependence among their components, a failure can range from affecting the normal operation of the unit to potentially triggering a chain reaction or even leading to a major accident. Therefore, researching a hydropower generating unit operation and maintenance model based on "condition-based maintenance" is of great significance for accurately assessing the operating status of hydropower generating units and detecting signs of deterioration and failure at an early stage.

[0003] However, existing methods still have the following shortcomings: First, they only focus on the temporal or spatial relationship between pulsating signals and operating parameters, ignoring the complex interactions between these relationships. Second, simply summing the averages of spatiotemporal relationships ignores the actual data distribution, which may lead to the loss of valuable information. Furthermore, summing the averages from various signal degradation state indicators may dilute the characteristics of key degradation information, potentially impairing the accuracy of performance assessment. Therefore, this invention proposes a physical-spatiotemporal weighted map construction method for assessing the health status of mixed-flow turbine units. Summary of the Invention

[0004] The main objective of this invention is to address the problem in existing technologies that use graph data to model pulsating signals and operating parameters, neglecting the inherent connections between spatial and characteristic relationships, thereby affecting the evaluation results.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] According to a first aspect of the present invention, the present invention claims protection for a method for assessing the health status of a mixed-flow turbine unit based on a physical spatiotemporal weighted graph, comprising the steps of:

[0007] S1, Draw the geometric model of each mixed-flow turbine in the mixed-flow turbine unit to obtain the grid node pressure distribution of the mixed-flow turbine, and use actual pulsating signal data to construct a high-fidelity model characterizing the actual situation of the mixed-flow turbine;

[0008] S2, use graph data to obtain the relationship between the operating parameters and pulsating signals of the mixed-flow turbine, and construct physical space edge connection relationships and characteristic edge connection relationships based on the spatial coordinates and characteristic data of the pulsating signals;

[0009] S3, weights are assigned to the physical space edge connection relationship and the feature edge connection relationship using a weight calculation function and weighted fusion is performed to generate a physical spatiotemporal weighted graph. The physical spatiotemporal weighted graph is then input into a graph convolution model for training to obtain a health state model of the mixed-flow turbine.

[0010] S4. Input the actual data of the mixed-flow turbine into the health status model, output the health label value of the actual observed signal, calculate the deterioration status index through the health label value and the actual pulsation signal, and evaluate the health status of the mixed-flow turbine through the determined health index.

[0011] Furthermore, step S1 also includes:

[0012] Draw a geometric model of the same scale based on the mixed-flow turbine drawings provided by the manufacturer;

[0013] The geometric model is meshed, and the pressure distribution of the mesh nodes of the mixed-flow turbine is simulated and obtained by computational fluid dynamics methods.

[0014] The pressure distribution of grid nodes is corrected using actual pulsating signal data to construct a high-fidelity model characterizing the actual situation of the mixed-flow turbine. Data augmentation is then performed based on the data generated by this high-fidelity model to obtain a high-fidelity pulsating signal dataset containing information on the mechanism of the mixed-flow turbine unit.

[0015] Furthermore, the method also includes:

[0016] The components of the geometric model described in step S1 include at least:

[0017] The components include the volute, double-row blades, impeller, and tailrace. The impeller was modeled using BladeGen with parametric blade modeling. By setting parameters such as the meridional plane coordinates of the flow channel, the blade wrap angle, and the cross-sectional thickness curve, a high-precision blade model was automatically generated. The other components were modeled using UG.

[0018] The computational fluid dynamics method described above uses the inlet section of the volute as the total pressure inlet boundary condition and the outlet section of the tailrace pipe as the static pressure outlet boundary condition. It uses the SST k-ω turbulence model to simulate the flow and obtain pressure pulsation signals.

[0019] Furthermore, step S2, which constructs a representation of physical space edge connectivity and feature edge connectivity based on the spatial coordinates and feature data of the pulsating signal, further includes:

[0020] Normalization and feature extraction operations are performed on the pulsating signal and operating parameters, and Min-Max normalization is applied for data normalization processing.

[0021] Principal component analysis and kernel principal component analysis are used to project the original signal into a high-dimensional feature space for feature extraction, and the feature representation is calculated. The Min-Max calculation formula is as follows:

[0022]

[0023] The method for calculating the feature edge connection relationship uses cosine similarity to measure the feature similarity between nodes. It connects a node to its k closest pulsating signal nodes in terms of features with edges to uncover the correlation between features. The cosine similarity calculation formula is as follows:

[0024]

[0025] The physical space edge connection calculation method uses Euclidean distance to measure the spatial distance between nodes based on the spatial coordinates and relative positions of the pulsating signal nodes. It connects each node with its k closest pulsating signal nodes in space using edges to explore the spatial correlation of pulsating signals.

[0026] The formula for calculating Euclidean distance is as follows:

[0027]

[0028] Furthermore, step S3 also includes:

[0029] When calculating the weighted edge connection relationship, weights are assigned to pulsating signal nodes by calculating the variance of spatial coordinates and eigenvalues, thereby strengthening the correlation between nodes that are spatially close.

[0030] By weighting the characteristic edge connection relationships and the spatial edge connection relationships, the final physical space weighted edge connection relationships are obtained;

[0031] The convolutional model for the physical-temporal weighted graph input consists of a three-layer Chebyshev graph convolutional network, a one-layer bidirectional long short-term memory network, and two fully connected layers.

[0032] The Chebyshev graph convolutional network is used to mine the spatial correlation of nodes in the constructed physical-temporal weighted graph, and the bidirectional long short-term memory network is used to mine the temporal correlation of the physical-temporal weighted graph at different time periods.

[0033] Furthermore, step S4 also includes:

[0034] The degradation status index is calculated by extracting key features of the health label value and the actual pulsation signal through principal component analysis.

[0035] Calculate the Euclidean distance between the key features of the extracted health tag values ​​and the key features of the actual pulsation signal, and construct a degradation state assessment index that conforms to the actual state.

[0036] The health indicators are based on 3 σ The principle is established by defining the health indicator by calculating the sum of the mean and standard deviation of the degradation status assessment index;

[0037] When the degradation status assessment index exceeds the defined health index, the turbine is determined to have deteriorated and requires maintenance.

[0038] This invention relates to the field of health status assessment of hydro-turbine units, and specifically to a method for assessing the health status of mixed-flow hydro-turbine units based on a physical-spatiotemporal weighted graph. The method involves constructing a geometric model of the mixed-flow hydro-turbine unit and calculating simulated pulsation signals. Based on the spatial coordinates and feature data of the pressure pulsation signals, a feature matrix and an adjacency matrix are constructed to capture physical spatial relationships and basic features. Temporal statistical parameters are used to assign weights to the feature relationships and physical spatial relationships. These relationships are then fused to generate a physical-spatiotemporal weighted graph, which is input into a graph convolution model to obtain the health status model of the mixed-flow hydro-turbine unit. Finally, the Euclidean distance between the predicted label and the actual observed signal under degraded conditions is calculated to derive a degradation status index to assess the health status of the mixed-flow hydro-turbine unit. This invention improves the level of health status assessment for mixed-flow hydro-turbine units by integrating prior knowledge of the mixed-flow hydro-turbine unit and combining it with a weighted graph representation learning method. Attached Figure Description

[0039] Figure 1 A flowchart illustrating the workflow of a health status assessment method for a mixed-flow turbine unit based on a physical spatiotemporal weighted graph, as claimed in an embodiment of the present invention.

[0040] Figure 2 A schematic diagram of a physical spatiotemporal weighted graph construction method for a health status assessment method of a mixed-flow turbine unit based on a physical spatiotemporal weighted graph, as claimed in an embodiment of the present invention;

[0041] Figure 3 A schematic diagram of the graph neural network state recognition model structure for a health status assessment method for a mixed-flow turbine unit based on a physical spatiotemporal weighted graph, as claimed in an embodiment of the present invention;

[0042] Figure 4 The accuracy of the turbine health status assessment task model identification for a mixed-flow turbine unit based on a physical spatiotemporal weighted graph, as claimed in this embodiment of the invention;

[0043] Figure 5 The diagram shows the health status assessment results of a mixed-flow turbine unit health status assessment method based on a physical spatiotemporal weighted graph, as claimed in an embodiment of the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0045] The terms "first," "second," and "third" used in this invention are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this invention are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0046] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0047] With the development of big data and artificial intelligence technologies, data-driven methods for assessing the health status of hydro turbines have rapidly emerged. These methods establish health status models between operating parameters and pulsating signals through machine learning or deep learning, and construct deviation indicators between measured data and the health status, thereby quantifying the unit's health condition. However, health status models built using machine learning methods struggle to reveal the implicit relationships between operating parameters and pulsating signals. To address this issue, graph theory has been introduced into unit condition assessment. Graph structures, through their unique spatial and topological properties, transform signals into graph data composed of nodes and edges. Graph Convolutional Networks (GCNs), by performing convolution operations on graphs, can extract signal features from discrete spatial structures.

[0048] According to a first embodiment of the present invention, the present invention claims protection for a method for assessing the health status of a mixed-flow turbine unit based on a physical spatiotemporal weighted graph, comprising the following steps:

[0049] S1, Draw the geometric model of each mixed-flow turbine in the mixed-flow turbine unit to obtain the grid node pressure distribution of the mixed-flow turbine, and use actual pulsating signal data to construct a high-fidelity model characterizing the actual situation of the mixed-flow turbine;

[0050] S2, use graph data to obtain the relationship between the operating parameters and pulsating signals of the mixed-flow turbine, and construct physical space edge connection relationships and characteristic edge connection relationships based on the spatial coordinates and characteristic data of the pulsating signals;

[0051] S3, weights are assigned to the physical space edge connection relationship and the feature edge connection relationship using a weight calculation function and weighted fusion is performed to generate a physical spatiotemporal weighted graph. The physical spatiotemporal weighted graph is then input into a graph convolution model for training to obtain a health state model of the mixed-flow turbine.

[0052] S4. Input the actual data of the mixed-flow turbine into the health status model, output the health label value of the actual observed signal, calculate the deterioration status index through the health label value and the actual pulsation signal, and evaluate the health status of the mixed-flow turbine through the determined health index.

[0053] Furthermore, step S1 also includes:

[0054] Draw a geometric model of the same scale based on the mixed-flow turbine drawings provided by the manufacturer;

[0055] The geometric model is meshed, and the pressure distribution of the mesh nodes of the mixed-flow turbine is simulated and obtained by computational fluid dynamics methods.

[0056] The pressure distribution of grid nodes is corrected using actual pulsating signal data to construct a high-fidelity model characterizing the actual situation of the mixed-flow turbine. Data augmentation is then performed based on the data generated by this high-fidelity model to obtain a high-fidelity pulsating signal dataset containing information on the mechanism of the mixed-flow turbine unit.

[0057] Furthermore, the method also includes:

[0058] The components of the geometric model described in step S1 include at least:

[0059] The components include the volute, double-row blades, impeller, and tailrace. The impeller was modeled using BladeGen with parametric blade modeling. By setting parameters such as the meridional plane coordinates of the flow channel, the blade wrap angle, and the cross-sectional thickness curve, a high-precision blade model was automatically generated. The other components were modeled using UG.

[0060] The computational fluid dynamics method described above uses the inlet section of the volute as the total pressure inlet boundary condition and the outlet section of the tailrace pipe as the static pressure outlet boundary condition. It uses the SST k-ω turbulence model to simulate the flow and obtain pressure pulsation signals.

[0061] In this embodiment, the reliability of the high-fidelity model is determined by comparing the pulsation signal output by the simulation with the actual measured pulsation signal. When the error between the two is less than 5%, the simulation result is considered reliable. If the error exceeds 5%, the boundary conditions of the model need to be adjusted, the simulation re-run, and this process repeated until the relative error drops below 5%.

[0062] Furthermore, step S2, which constructs a representation of physical space edge connectivity and feature edge connectivity based on the spatial coordinates and feature data of the pulsating signal, further includes:

[0063] Normalization and feature extraction operations are performed on the pulsating signal and operating parameters, and Min-Max normalization is applied for data normalization processing.

[0064] Principal component analysis and kernel principal component analysis are used to project the original signal into a high-dimensional feature space for feature extraction, and the feature representation is calculated. The Min-Max calculation formula is as follows:

[0065]

[0066] The method for calculating the feature edge connection relationship uses cosine similarity to measure the feature similarity between nodes. It connects a node to its k closest pulsating signal nodes in terms of features with edges to uncover the correlation between features. The cosine similarity calculation formula is as follows:

[0067]

[0068] The physical space edge connection calculation method uses Euclidean distance to measure the spatial distance between nodes based on the spatial coordinates and relative positions of the pulsating signal nodes. It connects each node with its k closest pulsating signal nodes in space using edges to explore the spatial correlation of pulsating signals.

[0069] The formula for calculating Euclidean distance is as follows:

[0070]

[0071] Reference Figure 2 This is a schematic diagram of the physical spatiotemporal weighted graph construction method in an embodiment of the present invention. By mapping the working condition parameters, including information such as water head, and the pulsating signal nodes to the node spatial coordinates, the node spatial attribute matrix is ​​obtained through analysis. The physical space weight vector is obtained based on variance calculation, and the physical space weighted edge connection relationship is obtained based on Euclidean distance calculation.

[0072] Simultaneously, the feature matrix is ​​obtained by processing the operating parameters and pulsating signal nodes, the node weight vector is obtained by calculating the variance, and the weighted feature matrix is ​​obtained by further analysis.

[0073] The weighted edge matrix is ​​obtained by fusing the weighted edge connection relationship and the weighted feature matrix in the physical space, and a weighted spatiotemporal graph in the physical space is obtained based on the weighted edge matrix.

[0074] Furthermore, step S3 also includes:

[0075] When calculating the weighted edge connection relationship, weights are assigned to pulsating signal nodes by calculating the variance of spatial coordinates and eigenvalues, thereby strengthening the correlation between nodes that are spatially close.

[0076] By weighting the characteristic edge connection relationships and the spatial edge connection relationships, the final physical space weighted edge connection relationships are obtained;

[0077] Reference Figure 3 The convolutional model for the physical-temporal weighted graph input consists of a three-layer Chebyshev graph convolutional network, a one-layer bidirectional long short-term memory network, and two fully connected layers.

[0078] The Chebyshev graph convolutional network is used to mine the spatial correlation of nodes in the constructed physical-temporal weighted graph, and the bidirectional long short-term memory network is used to mine the temporal correlation of the physical-temporal weighted graph at different time periods.

[0079] Furthermore, step S4 also includes:

[0080] The degradation status index is calculated by extracting key features of the health label value and the actual pulsation signal through principal component analysis.

[0081] Calculate the Euclidean distance between the key features of the extracted health tag values ​​and the key features of the actual pulsation signal, and construct a degradation state assessment index that conforms to the actual state.

[0082] The health indicators are based on 3 σ The principle is established by defining the health indicator by calculating the sum of the mean and standard deviation of the degradation status assessment index;

[0083] When the degradation status assessment index exceeds the defined health index, the turbine is determined to have deteriorated and requires maintenance.

[0084] This invention focuses on a mixed-flow turbine unit at a hydroelectric power station. Sensors are installed on key overflow components, including the volute, guide vanes, flow channels, and guide pipes, to collect multi-source monitoring signals. Data acquisition covers 363 days of multi-source time-series data from January 20, 2022 to January 18, 2023. This dataset includes monitoring signals and operating parameters such as head, power, and flow rate. The main technical parameters of the mixed-flow turbine unit are shown in Table 1.

[0085] Table 1

[0086]

[0087] To enhance the diversity of the dataset, twin data was generated using the simulation model, thereby expanding the entire dataset. First, a geometric model was established based on the geometric parameters of the mixed-flow turbine unit. The geometric model was then imported into TurboGrid for mesh generation. The mesh model was then imported into ANSYS FLUENT, and computational fluid dynamics techniques were used to analyze the internal flow field distribution of the flow components of the mixed-flow turbine unit. After verifying the high fidelity of the model, simulation data was generated to expand the dataset.

[0088] The health status model was constructed using operating parameters and pulsating signals from January 20, 2022 to June 15, 2022 as health status data. The pulsating dataset included 3457 samples from 10 monitoring points; 70% of these samples were allocated to the training set, while the remaining 30% were reserved for testing. Common operating parameters such as head, power, and flow rate were selected as inputs to the health status model. This invention combines node features with spatial relationships to construct a physical-spatiotemporal weighted graph. The constructed graph was then imported into ChebGCN-BiLSTM to train the health status model. Based on this, a degradation state index was calculated to quantify the degree of degradation of the mixed-flow turbine unit. The model structure is detailed in Table 2, and the learning rate was set to 0.0001.

[0089] Table 2

[0090]

[0091]

[0092] To verify the effectiveness of the present invention, refer to Figure 4This invention compares with four classic production line status recognition methods. The comparison methods include Support Vector Regression (SVR), Backpropagation Neural Network (BPNN), CNN-LSTM, and Multi-Head Attention Mechanism Network (MSNN). Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are used as evaluation metrics. Each model undergoes 10 comparative experiments, and the average value is taken as the final result. The study shows that all experiments demonstrate good fitting accuracy. Notably, the SVR model, which relies on machine learning, has the highest errors in MAE and RMSE, at 10.32% and 12.76%, respectively, indicating that simple linear models may not adequately capture the information required for multi-dimensional prediction. Among the various models tested, the ChebGCN-BiLSTM method exhibits superior fitting performance, with the lowest MAE and RMSE values ​​of 4.30% and 5.67%, respectively. By transforming the raw data into graphical structured data, mining the spatial relationships and feature information of pulsating signals can provide more information. This integration facilitates a thorough exploration of the intrinsic connections within time-series data, greatly enriching the representational capabilities of the original information. Subsequently, health status was assessed using the constructed health status model, with a total of 14,005 samples evaluated from June 15, 2022 to December 21, 2022. An evaluation cycle was conducted every hour, resulting in 2,600 evaluation cycles.

[0093] Reference Figure 5 This is a health status assessment result graph of a mixed-flow turbine unit based on a physical spatiotemporal weighted graph, as claimed in this invention. Although the curve fluctuates, the mechanical performance of the mixed-flow turbine unit gradually deteriorates over time due to the accumulation of damage, leading to a significant performance decline. This degradation is reflected in the corresponding increase in the deterioration state index values. The dashed line represents the warning threshold determined by the 3σ principle, which defines the standard by calculating the sum of the mean and standard deviation of the deterioration state index values. Typically, maintenance is required when the degradation state reaches a certain level to ensure that the deterioration state index recovers to a healthy state after intervention. It is worth noting that the provided data does not include the maintenance period; therefore, the observed increase in the deterioration state index accurately reflects the continuous degradation of the mixed-flow turbine unit, effectively confirming the effectiveness of the method.

[0094] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0095] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

[0096] The specific embodiments of the invention have been described in detail above, but these are merely examples, and the invention is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this invention. Therefore, all equivalent transformations, modifications, and improvements made without departing from the spirit and principles of this invention should be included within the scope of this invention.

Claims

1. A method for assessing the health status of a mixed-flow turbine unit based on a physical spatiotemporal weighted graph, characterized in that, Including the following steps: S1, Draw the geometric model of each mixed-flow turbine in the mixed-flow turbine unit, obtain the grid node pressure distribution of the mixed-flow turbine, and construct a high-fidelity model characterizing the actual situation of the mixed-flow turbine using actual pulsating signal data; S2, use graph data to obtain the relationship between the operating parameters and pulsating signals of the mixed-flow turbine, and construct physical space edge connection relationships and characteristic edge connection relationships based on the spatial coordinates and characteristic data of the pulsating signals; S3, weights are assigned to the physical space edge connection relationship and the feature edge connection relationship using a weight calculation function and weighted fusion is performed to generate a physical spatiotemporal weighted graph. The physical spatiotemporal weighted graph is then input into a graph convolution model for training to obtain a health state model of the mixed-flow turbine. S4, input the actual data of the mixed-flow turbine into the health status model, output the health label value of the actual observed signal, calculate the deterioration status index through the health label value and the actual pulsation signal, and evaluate the health status of the mixed-flow turbine through the determined health index; Step S1 further includes: Draw a geometric model of the same scale based on the mixed-flow turbine drawings provided by the manufacturer; The geometric model is meshed, and the pressure distribution of the mesh nodes of the mixed-flow turbine is simulated and obtained by computational fluid dynamics methods. The pressure distribution of grid nodes is corrected using actual pulsating signal data to construct a high-fidelity model that characterizes the actual situation of the mixed-flow turbine. Data augmentation is then performed based on the data generated by this high-fidelity model to obtain a high-fidelity pulsating signal dataset containing information on the mechanism of the mixed-flow turbine unit. Step S3 also includes: When calculating the weighted edge connection relationship, weights are assigned to pulsating signal nodes by calculating the variance of spatial coordinates and eigenvalues, thereby strengthening the correlation between nodes that are spatially close. By weighting the characteristic edge connection relationships and the spatial edge connection relationships, the final physical space weighted edge connection relationships are obtained; The convolutional model for the physical-temporal weighted graph input consists of a three-layer Chebyshev graph convolutional network, a one-layer bidirectional long short-term memory network, and two fully connected layers. The Chebyshev graph convolutional network is used to mine the spatial correlation of nodes in the constructed physical-temporal weighted graph, and the bidirectional long short-term memory network is used to mine the temporal correlation of the physical-temporal weighted graph at different time periods.

2. The method for assessing the health status of a mixed-flow turbine unit based on a physical spatiotemporal weighted graph as described in claim 1, characterized in that, Also includes: The components of the geometric model described in step S1 include at least: The components include the volute, double-row blade cascade, runner, and draft tube. The runner was modeled using BladeGen for parametric blade modeling. By setting parameters such as the flow channel meridional plane coordinates, blade wrap angle, and cross-sectional thickness curve, a high-precision blade model was automatically generated. The other components were modeled using UG. The computational fluid dynamics method described above uses the volute inlet section as the total pressure inlet boundary condition and the tailrace outlet section as the static pressure outlet boundary condition, employing SST (Sequencing and Stress Testing). Turbulence models are used to simulate flow and obtain pressure pulsation signals.

3. The method for assessing the health status of a mixed-flow turbine unit based on a physical spatiotemporal weighted graph as described in claim 1, characterized in that, Step S4 also includes: The degradation status index is calculated by extracting key features of the health label value and the actual pulsation signal through principal component analysis. Calculate the Euclidean distance between the key features of the extracted health tag values ​​and the key features of the actual pulsation signal, and construct a degradation state assessment index that conforms to the actual state. The health indicators are based on 3 The principle is established by defining the health indicator by calculating the sum of the mean and standard deviation of the degradation status assessment index; When the degradation status assessment index exceeds the defined health index, the turbine is determined to have deteriorated and requires maintenance.

Images