Gas turbine cascade pressure field sparse sensing and testing method based on sparse representation

Abstract

The invention discloses a gas turbine cascade pressure field sparse sensing and testing method based on sparse representation, and the method comprises the steps: constructing a three-dimensional numerical model according to a gas turbine cascade geometric model, and setting a corresponding boundary condition; performing unsteady numerical calculation by using CFD software, and outputting unsteady pressure field data; taking the unsteady pressure field data as a sample set, and learning from a given sample set through a dictionary learning method of the sample set to obtain a dictionary; sparse representation of the flow field is carried out by using the dictionary; dividing the flow field area into a plurality of clusters by adopting a K-means clustering method, and selecting sparse pressure measuring points in each cluster; and performing data reconstruction on the corresponding local flow field according to the pressure value of each sparse pressure measuring point, and superposing the local flow field of each cluster to realize the reconstruction of the global flow field. According to the method, sparse measured data and simulation data in the cascade can be cooperatively utilized, and high-precision flow field reconstruction in a whole-field state can be carried out only by relying on rare sensor information under a noise background.

Description

Technical Field

[0001] The present invention relates to the technical field of gas turbine flow field testing, and more particularly to a sparse representation-based gas turbine blade cascade pressure field sparse sensing and testing method. Background Art

[0002] As a strategic equipment, the safety of gas turbines is directly related to national defense security. At the same time, it is also a cutting-edge product in the field of energy and power equipment. Among them, turbine blades usually operate in extreme working environments of high temperature and high pressure. In this environment, turbine blades have a variety of secondary flow phenomena such as tip gap leakage vortex and channel vortex, and are also affected by flow structures such as upstream periodic wakes. Therefore, the flow situation inside the turbine becomes extremely complex and difficult to predict.

[0003] In existing measurement technologies, in order to accurately capture the details of the blade pressure field, a large number of measurement points are usually required to form a measurement matrix. First, a large number of measurement points require more hardware equipment such as pressure sensors, which increases the test cost and makes the test system complex and large, which is not conducive to installation, debugging and maintenance in practical applications. Secondly, the increase in the number of measurement points will lead to a corresponding increase in the difficulty of data processing and analysis. A large amount of raw data needs to be effectively stored, transmitted and processed, which puts higher requirements on the performance of data processing equipment, and may also prolong the data processing time, thereby affecting the real-time and efficiency of the test. Finally, too many measurement points may introduce measurement errors. In the actual test process, due to various factors such as sensor accuracy and environmental noise, differences and noise may occur between the measurement point data. These differences and noise may affect the accuracy of pressure field reconstruction and interfere with subsequent analysis and decision-making.

[0004] In recent years, the emerging intelligent method of sparse information perception has made it possible to perceive the whole field information with only a small number of sensors. The realization of this method mainly relies on the coordinated drive of measured data and numerical calculation. It mainly involves data-driven modal decomposition methods and artificial intelligence neural network algorithms. However, the current domestic research and application are mainly limited to classic external flow problems such as flow around cylinders and wings. There have been a few studies abroad on flow reconstruction for turbomachinery, but there are no relevant research reports on the sparse perception of blades, especially in the background of noise, the influence of extremely low number of sensors and noise intensity on reconstruction accuracy.

[0005] In summary, while ensuring the spatial resolution of the pressure field, the existing testing technology also faces challenges such as high testing cost, complex data processing, and measurement errors caused by noise background. Therefore, it is particularly important to develop a more efficient and accurate pressure field reconstruction method to make up for the shortcomings of the existing technology and achieve sparse reconstruction under noise background. Summary of the invention

[0006] In view of this, the present invention provides a sparse perception and testing method for the gas turbine blade pressure field based on sparse representation. In the face of the measurement and reconstruction of complex flow conditions inside the gas turbine blade, it can solve the problems of high cost, complex data processing and measurement errors caused by noise background in existing testing technologies.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] The present invention provides a sparse perception and testing method for a gas turbine blade cascade pressure field based on sparse representation, comprising the following steps:

[0009] S10. Construct a three-dimensional numerical model according to the geometric model of the gas turbine blade grid and set corresponding boundary conditions; perform unsteady numerical calculations using CFD software and output unsteady pressure field data;

[0010] S20, taking the unsteady pressure field data as a sample set, learning a dictionary from a given sample set by a dictionary learning method of the sample set; and performing sparse representation of the flow field by using the dictionary;

[0011] S30, using the K-means clustering method to divide the flow field area into multiple clusters, and selecting sparse pressure measurement points in each cluster;

[0012] S40, reconstructing data of the corresponding local flow field according to the pressure value of each sparse pressure measuring point, and superimposing the local flow fields of each cluster to achieve reconstruction of the global flow field.

[0013] Furthermore, in step S10, corresponding boundary conditions are set, including:

[0014] The velocity and temperature parameters of the airflow are set at the inlet section; the incoming flow velocity of the inlet section is set based on the Reynolds number; and the outlet section is set to a fixed static pressure condition to ensure the stable discharge of the simulated airflow.

[0015] Furthermore, the sparse representation in step S20 is achieved by a linear combination of an overcomplete dictionary and a sparse vector.

[0016] Furthermore, the step S20 further includes: setting an error tolerance;

[0017] The dictionary is used to perform sparse representation of the flow field, and the formula is as follows:

[0018]

[0019] In the formula, represents a sparse vector; S k is the sparse representation coefficient; y krepresents the sparse sampling value in the flow field; P k represents the measurement matrix; Ψ k represents the dictionary; ε represents the error tolerance.

[0020] Furthermore, the step S30 includes:

[0021] Randomly select K data points from a given flow field sample set as the initial clustering centers, denoted as μ 1 , μ 2 , …μ K ;

[0022] For each sample point x i , calculate its relationship with each cluster center μ k The distance of the sample point x i Assigned to the cluster C corresponding to the cluster center closest to it k ;

[0023] For each cluster C k , recalculate the mean of all sample points in the cluster and use this mean as the new cluster center:

[0024] Repeat the above allocation and update steps until the cluster center position no longer changes significantly or reaches the preset number of iterations, and the clustering process converges;

[0025] After the K-means algorithm, the flow field area is divided into K clusters, each cluster C k The flow field data within have similar flow characteristics and pressure distribution;

[0026] One or more pressure measurement points are randomly selected in each cluster as representative measurement points.

[0027] It can be seen from the above technical solutions that compared with the prior art, the present invention has the following technical advantages:

[0028] This method can make use of the sparse measured data in the blade cascade in conjunction with the simulation data. It can reconstruct the flow field with high precision in the whole field state by relying only on the sparse sensor information under the noise background. In addition, this method still shows high robustness in the sparse reconstruction of the flow field under high noise environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention 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 only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0030] Figure 1A flow chart of a sparse perception and testing method for a gas turbine blade cascade pressure field based on sparse representation provided in an embodiment of the present invention.

[0031] Figure 2 A schematic diagram of a sparse perception and testing method for a gas turbine blade cascade pressure field based on sparse representation provided in an embodiment of the present invention;

[0032] Figure 3 A three-dimensional numerical model diagram of a gas turbine blade cascade flow passage provided in an embodiment of the present invention;

[0033] Figure 4 The flow field segmentation and measurement point distribution diagram based on K-means clustering analysis provided in the embodiment of the present invention;

[0034] Figure 5 A schematic diagram of local reconstruction of a clustered and block pulsating flow field provided by an embodiment of the present invention;

[0035] Figure 6 A sparse reconstruction image of a pulsating flow field under a noisy background provided by an embodiment of the present invention;

[0036] Figure 7 A comparison diagram of the global reconstruction and local reconstruction errors of sparse representation under a noisy background provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] When studying the fluid mechanics of turbomachinery in gas turbines, even simple aerodynamic shapes will produce complex flows with extensive spatiotemporal characteristics. Due to limited sensor data and the fact that high-precision models of the flow field are limited by grid scale and time step, it is difficult to directly reconstruct the flow field of the inner channel of the gas turbine. Intelligent sensing technology is regarded as an important means to overcome the limitations of traditional experimental and simulation sensing methods. Therefore, the example of the present invention incorporates known sparse sampling information into the flow field reconstruction process based on sparse representation, in order to achieve high-precision reconstruction of the flow field while reducing the amount of data and enhancing the interpretability of the gas turbine flow characteristics.

[0039] Reference Figure 1-2 As shown, an embodiment of the present invention discloses a sparse perception and testing method for a gas turbine blade cascade pressure field based on sparse representation, comprising the following steps:

[0040] S10. Construct a three-dimensional numerical model according to the geometric model of the gas turbine blade grid and set corresponding boundary conditions; perform unsteady numerical calculations using CFD software and output unsteady pressure field data;

[0041] S20, taking the unsteady pressure field data as a sample set, learning a dictionary from a given sample set by a dictionary learning method of the sample set; and performing sparse representation of the flow field by using the dictionary;

[0042] S30, using the K-means clustering method to divide the flow field area into multiple clusters, and selecting sparse pressure measurement points in each cluster;

[0043] S40, reconstructing data of the corresponding local flow field according to the pressure value of each sparse pressure measuring point, and superimposing the local flow fields of each cluster to achieve reconstruction of the global flow field.

[0044] In this embodiment, by constructing a three-dimensional numerical model of the gas turbine blade grid, extracting the pressure field data inside the blade grid, and reconstructing the flow field using the sparse representation theory, the computational complexity and hardware requirements are reduced. In addition, the K-means clustering algorithm is used to divide the flow field area into different clusters, and the data similarity within each cluster is high, so a small number of sparse measurement points are selected in each cluster to reconstruct the local flow field, and finally, high-precision reconstruction of the global flow field is achieved by superimposing all local flow fields. The number of sensors is reduced by sparse reconstruction, and high-precision perception of the complex flow field of the gas turbine blade grid is achieved. Therefore, through this method, the pressure field of the gas turbine blade grid can be monitored and analyzed in a low-cost and high-precision manner. This technology is very helpful for improving the efficiency, safety and maintenance capabilities of gas turbines.

[0045] The following is a detailed description of each of the above steps:

[0046] In step S10:

[0047] Based on the geometric model of the gas turbine blade row, the three-dimensional flow field model is constructed using CFD (computational fluid dynamics) software. Figure 3 The figure shows a numerical three-dimensional flow field model of a gas turbine cascade. In the simulation process, the inlet section is first set up, and the periodic wake generated by the upstream blades in the channel is simulated by the periodic vortex generated by the micro-cylindrical boundary layer separation. The inlet flow velocity of the computational domain is set based on the Reynolds number, and the boundary condition at the outlet is set to a fixed static pressure.

[0048] That is, in this model, appropriate boundary conditions (such as inlet velocity and temperature, outlet pressure, etc.) are set and unsteady numerical calculations are performed. The result of the calculation is unsteady pressure field data, that is, the pressure distribution at certain specific locations in the cascade flow channel that changes with time.

[0049] After setting the boundary conditions, the CFD software is used to perform unsteady calculations on the model. Unsteady calculations mean that the flow field changes over time and is not static. This calculation outputs the pressure changes in the flow field by simulating different time steps. The solution time step is precisely set to 0.001 milliseconds, and the calculation cycle is divided into multiple dimensionless moments to observe the dynamic changes of the airflow over time.

[0050] After CFD simulation calculation, the unsteady pressure field data inside the entire blade flow channel can be finally obtained. These data include pressure values ​​at different positions and at different times in the flow channel, forming a pressure distribution diagram that changes with time. These pressure field data will serve as a sample set for subsequent analysis and reconstruction.

[0051] Specifically, the solution time step of the simulation model is precisely set to 0.001ms. A complete wake passing cycle is defined as 200 dimensionless moments, where the interval between two adjacent dimensionless moments t * =Δt / t is defined as the ratio of the adjacent data storage time to the trail passing period. In view of the dominant time mean characteristic of the flow field, in the process of sparse reconstruction of the gas turbine blade flow field, the present invention example pays special attention to the reconstruction effect of the pulsating flow field at an unknown time. The pulsating flow field refers to the part of the flow field that is not averaged over time, and usually represents more complex and transient changes in the flow field. The invention pays special attention to how to reconstruct such a pulsating flow field at an unknown time within a given limited time period.

[0052] Taking the blade section with 50% blade height as an example, the flow field sample set Contains pulsating flow field pressure data of 160 dimensionless moments (unsteady pressure field data contains flow field pressure values ​​at different moments. Because the flow field changes continuously with time, the pressure field data can be divided into time average value and pulsating value outside the mean. By processing the pulsating flow field data, more obvious flow field changes can be observed. The pulsating flow field pressure data here is part of the unsteady pressure field data. The process of reconstructing the flow field is to reconstruct the pulsating flow field at the unknown moment, and then add it to the time average value of the flow field in the sample set to obtain an approximate value of the flow field at the unknown moment). Each column represents the pressure value of a discrete point in the plane space of the pulsating flow field.

[0053] In step S20:

[0054] The unsteady pressure field data obtained by CFD calculation is used as a sample set, and a sparse representation dictionary is generated through the dictionary learning method. This dictionary can be considered to be composed of multiple sets of feature vectors, which can represent the main characteristics of the flow field. Then, these feature vectors and sparse coefficients are used to approximate the flow field data at each moment. The generated dictionary Ψ k, used to sparsely represent the pressure field data at each moment: Represents a sparse vector.

[0055] There are low-dimensional structures in high-dimensional flow fields that can dominate flow characteristics, which provides potential possibilities for flow field reconstruction using sparse representation. It is calculated by the numerical model, where x represents the pressure value of all discrete points in the cascade flow field at a certain moment, and n represents the number of discrete points. * =161~170 flow field data is used as the test set. All sparse sampling values ​​are from the test set to verify the accuracy of the sparse perception of the flow field pressure value at unknown time. Flow field test set The superscript represents the time.

[0056] This sparse representation model can reconstruct the global situation of complex flow fields with fewer measurement points and data, and significantly reduce computational costs and data storage requirements.

[0057] The core of this process is to obtain a dictionary that can capture the flow field characteristics through dictionary learning, and then use a small number of sparse coefficients to reconstruct the entire flow field data through the sparse representation method, and set a reasonable error tolerance to ensure the accuracy and reliability of the reconstruction.

[0058] In step S30, it specifically includes:

[0059] S301, randomly select K data points from a given flow field sample set as initial cluster centers, denoted as μ 1 , μ 2 , …μ K ;

[0060] S302, for each sample point xi, calculate its distance to each cluster center μk, and convert the sample point x i Assigned to the cluster C corresponding to the cluster center closest to it k ; The distance is usually calculated using the Euclidean distance: d(x i , μ k )=||x i -μ k || 2 ;

[0061] S303. For each cluster C k , recalculate the mean of all sample points in the cluster and use this mean as the new cluster center:

[0062] S304, repeat the above allocation and update steps (steps S302 and S303) until the cluster center position no longer changes significantly or reaches a preset number of iterations, and the clustering process converges;

[0063] S305, after the K-means algorithm, the flow field area is divided into K clusters, each cluster C k The flow field data within have similar flow characteristics and pressure distribution;

[0064] S306: Randomly select one or more pressure measuring points in each cluster as representative measuring points.

[0065] In step S30, in order to reduce the computational complexity of flow field reconstruction, the flow field area is divided into several clusters according to its flow characteristics, and the flow characteristics in each cluster are similar. Using the K-means algorithm, the pressure field data is divided into K clusters according to the flow characteristics. Sparse measurement points are selected in each cluster for data acquisition and reconstruction. The center point of each cluster and its sparse measurement point position are obtained for subsequent local flow field reconstruction.

[0066] The flow field sample set itself represents the pressure values ​​at different times in the flow field. According to the K-means clustering algorithm, the flow field sample set is divided into k different clusters X according to the pressure values. k , each cluster represents a type of the same flow feature in the flow field. The flow field data within the same cluster will show a high degree of similarity, while the data between different clusters will show a lower degree of similarity. This feature makes it possible to effectively distinguish different areas in the flow field based on the clustering results, thereby providing strong support for the subsequent sparse measurement point selection and local reconstruction of the flow field.

[0067] After partitioning the flow field based on the sample set, a measurement point is randomly selected in each area. All sparse sampling values ​​are selected from the test set. Except for the sparse sampling values, the flow field data at all times in the test set is equivalent to unknown. For the unknown time flow field to be reconstructed, in each cluster, y k =P k x k Among them, y k represents the sparse sampling value in each cluster, which is known; P k As the measurement matrix, it represents the distribution of the measurement points in each cluster, which is also known; k Represents the full flow field data at an unknown time, which is the reconstruction target required by the example of the present invention.

[0068] In the actual sparse sampling process, data noise is inevitably introduced. In a noisy environment, the authenticity and consistency of the flow field data are destroyed, resulting in an increase in reconstruction errors. Therefore, additive noise is defined as: σ represents the noise intensity, and E is the energy average of the flow field. In the present example, the sparse sampling value in the flow field is y k Considering the data noise, as the noise intensity increases, y k It represents the sum of the true value of the pressure at the measuring point in the flow field and the additive noise, x i Represents the pressure value at all discrete positions in the flow field, and n represents the number of discrete positions. The sparse sampling values ​​of the examples of the present invention are based on the simulation model, and the additive noise is equivalent to randomly adding data noise that conforms to the standard normal distribution in the flow field. On this basis, sparse sampling is performed. Ensure that all sparse sampling values ​​contain noise.

[0069] In step S40:

[0070] Take the sample set of each cluster itself as the overcomplete dictionary X k =Ψ k , using sparse representation theory, Among them, S k is the sparse representation coefficient; ε represents the error tolerance. When setting the error tolerance of sparse representation, a larger error tolerance is easier to reconstruct the flow field but will increase the reconstruction error. Therefore, it is necessary to weigh the relationship between reconstruction accuracy and computational cost. In the example of the present invention, ε = 0.01. The flow field reconstruction problem is simplified to finding the sparse sampling value y k The sparsest vector that matches In getting the sparse vector Then, based on the sparse representation theory, the flow field at the unknown time in each cluster is: The local flow fields reconstructed in all the above clusters are linearly superimposed to obtain the global flow field of the cascade at the unknown time.

[0071] like Figure 4 As shown in the figure, the K-means clustering analysis algorithm is used to finely divide the gas turbine blade flow field data into nine clusters. Each cluster has unique flow characteristics and represents a specific pressure distribution and flow state in the flow field. In order to ensure the uniformity of the measurement point distribution, a measurement point is randomly selected in each cluster. By classifying the flow field areas with similar flow characteristics into the same cluster, the consistency of the flow mechanism within each cluster is ensured, thereby greatly simplifying the process of local flow field reconstruction. These different clusters form obvious partitions in the flow field. In particular, clusters 2 and 3 are mainly concentrated near the suction surface of the blade. Since the suction surface of the blade is often accompanied by various complex flow phenomena, dividing the flow field in these areas into a cluster separately helps to more accurately reconstruct and analyze the flow characteristics of the area.

[0072] Figure 5The local reconstruction results of the pulsating flow field after clustering and segmentation are shown. The selected measuring points in each cluster are used as the data source, and only the flow field data within the same cluster are independently reconstructed based on sparse representation. Although each cluster contains only one measuring point and provides limited flow field information, it is correspondingly less affected by noise. When constructing a complete flow field, the local reconstruction results of all clusters are superimposed. This approach is based on the global consistency and coherence of the flow field data. Since the local reconstruction of each cluster is based on its unique internal flow characteristics and the corresponding data source points, there may be slight differences in the reconstruction results between different clusters. However, by superimposing the local reconstruction results of all clusters, the example of the present invention can integrate these differences and finally construct a complete flow field that is consistent and coherent as a whole.

[0073] Figure 6 (a) shows the original data of the pulsating flow field, and Figure 6 (b) depicts the noise background and the global sampling point position distribution when the noise intensity is σ = 0.4. The sampled data are embedded in the noise background. Figure 6 (c) and Figure 6 (d) shows the flow field reconstruction results based on GappyPOD and sparse representation respectively. It can be observed from the figure that under the background of noise, the GappyPOD method finds the best function match for the data by minimizing the sum of squared errors. However, when there is noise in the data, outliers may be given too large weights, which may increase the sensitivity of the reconstruction model to noise, thereby increasing the error of flow field reconstruction. The flow field reconstruction is performed independently in each cluster, which effectively avoids the complexity and uncertainty that may be brought about by global reconstruction. Under the background of noise, the local reconstruction method of the flow field using sparse representation method and clustering can significantly reduce the interference of noise on data reconstruction, thereby providing more accurate and reliable flow field reconstruction results.

[0074] Figure 7The comparison of the average reconstruction error of the global sparse representation reconstruction of the flow field in the test set and the local sparse representation based on K-means clustering under different noise backgrounds under the same distribution of measuring points is shown. When the noise intensity reaches σ=0.6, the reconstruction error of the global sparse representation increases by 5.63% compared with the ideal situation. With the further increase of noise intensity, especially in the area with complex flow, the influence of noise on the reconstruction error becomes more significant. However, after the flow field is processed by clustering and blocking, the reconstruction error based on local sparse representation only increases by 0.14% compared with the ideal situation under the same noise intensity. This shows that in each cluster processed by blocking, the flow field reconstruction shows low sensitivity to the increase of noise intensity. The results of the examples of the present invention show that under the background of high-intensity noise, especially in complex flow conditions, the local reconstruction method of the flow field based on clustering and blocking shows more significant advantages. This method effectively reduces the influence of noise on the reconstruction results and improves the accuracy and reliability of reconstruction by dividing the flow field into multiple clusters with similar flow characteristics and reconstructing the flow field independently in each cluster.

[0075] The present invention combines the numerical results of the impeller machinery flow field with sparse sampling information, and realizes the reconstruction of the global flow field based on sparse representation, thereby reducing the computational cost and providing strong data support for performance prediction and evaluation. In addition, the present invention proposes a measurement point arrangement and reconstruction strategy based on block clustering unsupervised learning. In view of the significant impact of the number of sensors and noise intensity in the blade grid on the perception accuracy under the background of noise, the present invention decomposes the global flow field reconstruction problem into the superposition processing of local flow field reconstruction, aiming to achieve the goal of high-precision reconstruction in the background of noise using a limited number of sensors, thereby realizing rapid estimation of the flow field.

[0076] By reconstructing the unsteady flow pressure field inside the blade, the present invention provides more accurate and comprehensive data support for the experiment, which helps to reveal the complex mechanism of internal flow. The present invention focuses on two research areas: sparse perception of blade state and sparse reconstruction of pressure field. On this basis, it can help to further build a "perception-recognition" technology system for unstable flow in gas turbine blade pressure field.

[0077] The method proposed in the present invention solves the problem that it is difficult to densely arrange measurement points in the gas turbine blade grid in the real operating environment. More importantly, through the local reconstruction method of the flow field based on K-means clustering and block, the classified flow field data helps to more deeply understand the physical characteristics and flow mechanism of the fluid in the gas turbine flow channel, and then determine the areas with the most drastic flow field changes and the most complex structures. At the same time, the flow field is locally reconstructed, thereby effectively avoiding the complexity and uncertainty that may be caused by global reconstruction, which further improves the robustness of the sparse reconstruction of the full flow field under the background of high noise intensity.

[0078] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0079] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A sparse perception and testing method for gas turbine blade cascade pressure field based on sparse representation, characterized in that: The following steps are involved: S10. Construct a three-dimensional numerical model according to the geometric model of the gas turbine blade grid and set corresponding boundary conditions; perform unsteady numerical calculations using CFD software and output unsteady pressure field data; S20, taking the unsteady pressure field data as a sample set, and learning a dictionary from a given sample set by a dictionary learning method of the sample set; Using the dictionary to perform sparse representation of the flow field; S30, using the K-means clustering method to divide the flow field area into multiple clusters, and selecting sparse pressure measurement points in each cluster; S40, reconstructing data of the corresponding local flow field according to the pressure value of each sparse pressure measuring point, and superimposing the local flow fields of each cluster to achieve reconstruction of the global flow field.

2. The sparse perception and testing method for gas turbine blade cascade pressure field based on sparse representation according to claim 1, characterized in that: In step S10, corresponding boundary conditions are set, including: The velocity and temperature parameters of the airflow are set at the inlet section; the incoming flow velocity of the inlet section is set based on the Reynolds number; and the outlet section is set to a fixed static pressure condition to ensure the stable discharge of the simulated airflow.

3. The sparse perception and testing method for gas turbine blade cascade pressure field based on sparse representation according to claim 1, characterized in that: The sparse representation in step S20 is achieved by a linear combination of an overcomplete dictionary and a sparse vector.

4. The sparse perception and testing method for gas turbine blade cascade pressure field based on sparse representation according to claim 1, characterized in that: The step S20 further includes: setting an error tolerance; The dictionary is used to perform sparse representation of the flow field, and the formula is as follows: In the formula, represents a sparse vector; S k is the sparse representation coefficient; y k represents the sparse sampling value in the flow field; P k represents the measurement matrix; Ψ k represents the dictionary; ε represents the error tolerance.

5. The sparse perception and testing method for gas turbine blade cascade pressure field based on sparse representation according to claim 1, characterized in that: The step S30 comprises: Randomly select K data points from a given flow field sample set as the initial cluster centers, denoted as μ1, μ2, …μ K ; For each sample point x i , calculate its relationship with each cluster center μ k The distance of the sample point x i Assigned to the cluster C corresponding to the cluster center closest to it k ; For each cluster C k , recalculate the mean of all sample points in the cluster and use this mean as the new cluster center: Repeat the above allocation and update steps until the cluster center position no longer changes significantly or reaches the preset number of iterations, and the clustering process converges; After the K-means algorithm, the flow field area is divided into K clusters, each cluster C k The flow field data within have similar flow characteristics and pressure distribution; One or more pressure measurement points are randomly selected in each cluster as representative measurement points.

Images