SVM-PSO-based ship gas turbine performance degradation identification method

Abstract

The invention relates to a ship gas turbine performance degradation identification method based on SVM-PSO. Measurable parameters of an inlet and an outlet of each component of the gas turbine are collected, a mechanism model of the ship gas turbine is combined, a specific component with performance degradation is positioned based on an SVM model, and then the degradation degree of the component with performance degradation is identified based on a PSO algorithm. The method can accurately identify the performance degradation state of the ship gas turbine under the condition that at most three parts are degraded at the same time. According to the method, by positioning the degraded part, the number of performance degradation factors needing to be identified is reduced, the search dimension of a performance degradation factor identification space is reduced, the precision of ship gas turbine performance degradation identification is improved, and the solving speed of the identification process is also improved.

Description

Technical Field

[0001] The invention relates to the field of fault diagnosis systems, in particular to fault diagnosis of gas turbines, and specifically to a method for identifying performance degradation of a ship gas turbine based on SVM-PSO. Background Art

[0002] In recent years, with the in-depth development of marine resources and the high attention paid to marine rights and interests by countries around the world, the technology in the field of ships has developed rapidly. Gas turbines are widely used in ship propulsion as a form of power with high power density, good maneuverability and fast response speed. As the operation time of gas turbines increases, the performance of multiple gas path components such as compressors and turbines will degrade due to the accumulation of pollutants and the weakening of component sealing. Identifying the components with degraded performance of gas turbines and the corresponding degree of degradation, and guiding the cleaning or maintenance of the corresponding components can reduce safety hazards and reduce operating costs.

[0003] For ship gas turbines, there are many compressor and turbine components that may experience performance degradation. Due to the coaxial mechanical connection and the series connection of the gas paths of the components, there is a strong coupling between the components, making it difficult to identify performance degradation. The existing performance degradation identification methods are mainly divided into two categories: model-based and data-driven. When the number of degradation factors is large, the optimization search space dimension of the model-based method is high, which not only significantly increases the solution time, but also easily causes the solution process to fall into the local optimum, and the accuracy is significantly reduced. In addition, the sample size of performance degradation data in actual engineering is small, the degradation and fault sample data set is relatively difficult to obtain, and the characteristic parameters of the performance degradation prediction model are unclear, so the use of data-driven methods is limited.

[0004] The present invention effectively combines data-driven and model-based methods, adopts a support vector machine (SVM) model to identify degraded components, and then adopts a particle swarm optimization (PSO) algorithm to identify the degree of degradation of the corresponding components, thereby effectively identifying the degradation state of a gas turbine system in which up to three components are degraded at the same time, which not only improves the solution efficiency of the identification process, but also improves the identification accuracy. Summary of the invention

[0005] The present invention proposes a method for identifying performance degradation of a ship gas turbine based on SVM-PSO, which can effectively identify the degradation state of a gas turbine system with up to three components degraded at the same time. Due to the use of a support vector machine combined with a particle swarm optimization algorithm, the efficiency and accuracy of the identification solution are improved.

[0006] To achieve the above objectives, the present invention adopts the following technical solutions: Firstly, the inlet and outlet parameters of gas turbine components are collected, and then the components with performance degradation are determined through the SVM model. Finally, the PSO algorithm is used to identify the degradation degree of the degraded components, thereby realizing efficient and accurate identification of gas turbine performance degradation.

[0007] Furthermore, the process of establishing the SVM model is as follows: first, the Newton-Raphson algorithm is used to combine the principles of thermodynamics and kinetics to establish a nonlinear mechanism model of the gas turbine; then, the flow degradation factor and efficiency degradation factor are implanted in some component modules of the nonlinear mechanism model of the gas turbine, and the combustion efficiency degradation factor is implanted in other component modules, and the range of the degradation factor is set to 0.95~1.0, and the response characteristics of the inlet and outlet parameters of the gas turbine components under different degradation factors are obtained. White noise is implanted in the obtained data, and labels are added to the data according to the set degradation factor. When the degradation factors of a certain component are all 1, it is in a non-degraded state and the label is set to 0. When the flow degradation factor or efficiency degradation factor of a certain component is less than 1, or the combustion efficiency degradation factor is less than 1, it is in a degraded state and the label is set to 1, thereby obtaining an SVM model data set; the SVM model data set is randomly divided, 80% is used as a training set and 20% is used as a data set, and the kernel function of the SVM model is set to a cubic polynomial kernel function, and the SVM model is trained, and corresponding SVM models are established for different components.

[0008] Furthermore, the inlet and outlet parameters of the gas turbine components under steady-state operation are combined into a one-dimensional vector as a feature vector, and the feature vector is input into the SVM model to determine whether the component has been degraded. If it is determined that the performance of one of the above components has been degraded, the degree of degradation of the degraded component needs to be identified.

[0009] Furthermore, based on the collected inlet and outlet parameters of the gas turbine components, the gas turbine mechanism model is verified and corrected.

[0010] Furthermore, the identification process of the PSO algorithm is as follows: the degradation factor of the component with performance degradation is taken as the parameter to be solved, the initial population of degradation factors is randomly generated, and then the nonlinear mechanism model of the gas turbine is substituted to calculate the response characteristics of the inlet and outlet parameters of the gas turbine components corresponding to the corresponding initial population, and the mean square error MSE of the calculated data and the measured inlet and outlet parameters of the gas turbine components is taken as the objective function, and the objective function value is calculated, which is expressed as follows: , in: , , Indicates the operating status parameters of different components, corner marks Indicates measured data, subscript Represents the calculated data, Indicates the number of import and export parameters; Then, it is determined whether the objective function value reaches the termination condition. If so, the degradation degree is output. If not, the degradation factor of the population is updated based on the PSO algorithm, and a cyclic iterative calculation is performed until the set iteration number is met to complete the identification of the degradation degree of the degraded component.

[0011] Furthermore, the PSO algorithm is optimized by introducing random factors in the iterative solution process to improve the global search capability of the algorithm. In the iterative optimization process, as the number of iterations increases, the inertia weight is adaptively adjusted. In the early stage of the iteration, a smaller inertia weight is used to improve the global search capability of the algorithm. When the number of iterations is close to the set value, a larger inertia weight is used to improve the convergence of the algorithm.

[0012] Furthermore, the inlet and outlet parameters of the gas turbine components include any one or more of the following: low-pressure compressor outlet temperature, low-pressure compressor outlet pressure, low-pressure compressor flow, low-pressure compressor speed, high-pressure compressor outlet temperature, high-pressure compressor outlet pressure, high-pressure compressor flow, high-pressure compressor speed, fuel flow, power turbine outlet temperature, power turbine outlet flow and power turbine speed of the gas turbine.

[0013] After adopting the above scheme, the beneficial effects of the present invention are: 1. By collecting the inlet and outlet parameters of the measurable interface of the gas turbine, the degradation of different components is judged based on the support vector machine (SVM) model, which reduces the difficulty of classifying degraded components and improves the accuracy of classification of degraded components. Compared with Gaussian Bayes and random forest, it has the best judgment accuracy of degraded components; 2. Based on the gas turbine mechanism model combined with the particle swarm optimization (PSO) algorithm, the corresponding degradation degree of degraded components can be identified. By improving the PSO algorithm, the global search capability of the algorithm can be improved, and the identification accuracy can be improved; 3. The SVM-PSO identification method is a method that first reduces the dimension of the solution space and then optimizes the solution based on the reduced dimension space. It can effectively improve the efficiency and accuracy of performance degradation identification solution of gas turbines including multiple components. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is the principle diagram of gas turbine performance degradation identification based on SVM-PSO.

[0015] Figure 2 It is a comparison of the accuracy of the present invention with Gaussian Bayes (GB) and Random Forest (RF) in identifying the degradation of low-pressure compressor; Figure 3It is a comparison of the accuracy of the present invention with Gaussian Bayes (GB) and Random Forest (RF) in identifying high-pressure compressor degradation; Figure 4 It is a comparison of the accuracy of the present invention with Gaussian Bayes (GB) and Random Forest (RF) in identifying high-pressure turbine degradation; Figure 5 It is a comparison of the accuracy of the present invention with Gaussian Bayes (GB) and Random Forest (RF) in identifying low-pressure turbine degradation; Figure 6 The present invention is compared with Gaussian Bayes (GB) and random forest (RF) in terms of accuracy in identifying power turbine degradation. DETAILED DESCRIPTION

[0016] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0017] The present invention provides a method for identifying performance degradation of a ship gas turbine propulsion system based on SVM-PSO. Figure 1 As shown, the following steps are included: First, the measurable inlet and outlet parameters of gas turbine components are collected. Then, the SVM model (Support Vector Machine, also known as support vector machine) is used to determine the components with performance degradation. Finally, the PSO algorithm (Particle Swarm Optimization, also known as particle swarm algorithm) is used to identify the degradation degree of the degraded components, thereby achieving efficient and accurate identification of gas turbine performance degradation.

[0018] Specifically, the inlet and outlet parameters of the gas turbine components include any one or more of the following: the low-pressure compressor outlet temperature, low-pressure compressor outlet pressure, low-pressure compressor flow, low-pressure compressor speed, high-pressure compressor outlet temperature, high-pressure compressor outlet pressure, high-pressure compressor flow, high-pressure compressor speed, fuel flow, power turbine outlet temperature, power turbine outlet flow and power turbine speed of the gas turbine.

[0019] This embodiment is described by taking all the above import and export parameters as an example.

[0020] Specifically, the process of establishing the SVM model is as follows: first, the Newton-Raphson algorithm is used to combine the principles of thermodynamics and kinetics to establish a nonlinear mechanism model of the gas turbine. Subsequently, based on the collected inlet and outlet parameters of the gas turbine components, the gas turbine mechanism model is continuously verified and corrected; then, flow degradation factors and efficiency degradation factors are implanted in some modules of the nonlinear mechanism model of the gas turbine, including the low-pressure compressor, high-pressure compressor, high-pressure turbine, low-pressure turbine and power turbine modules, and combustion efficiency degradation factors are implanted in other modules such as the combustion chamber module. The range of the degradation factor is set to 0.95~1.0. Under a certain specific operating condition, the response characteristics of the low-pressure compressor outlet temperature, low-pressure compressor outlet pressure, low-pressure compressor flow, low-pressure compressor speed, high-pressure compressor outlet temperature, high-pressure compressor outlet pressure, high-pressure compressor flow, high-pressure compressor speed, fuel flow, power turbine outlet temperature, power turbine outlet flow and power turbine speed of the gas turbine under different degradation factors are obtained based on the model calculation. White noise is implanted into the obtained data, and labels are added to the data according to the set degradation factor. When the flow degradation factor and efficiency degradation factor of a certain module are both 1, or the combustion efficiency degradation factor of the combustion chamber module is 1, it is in a non-degraded state, and the label is set to 0. When the flow degradation factor or efficiency degradation factor of a certain module is less than 1, or the combustion efficiency degradation factor of the combustion chamber module is less than 1, it is a degraded state, and the label is set to 1, thereby obtaining the SVM model data set; the SVM model data set is randomly divided, and 80% is used as The kernel function of the SVM model is set to the cubic polynomial kernel function, and the SVM models of different components of low-pressure compressor, high-pressure compressor, combustion chamber, high-pressure turbine, low-pressure turbine and power turbine are obtained through training. Therefore, the SVM models include: low-pressure compressor degradation identification SVM model, high-pressure compressor degradation identification SVM model, combustion chamber degradation identification SVM model, high-pressure turbine degradation identification SVM model, low-pressure turbine degradation identification SVM model and power turbine degradation identification SVM model.

[0021] The low-pressure compressor outlet temperature, low-pressure compressor outlet pressure, low-pressure compressor flow, low-pressure compressor speed, high-pressure compressor outlet temperature, high-pressure compressor outlet pressure, high-pressure compressor flow, high-pressure compressor speed, fuel flow, power turbine outlet temperature, power turbine outlet flow and power turbine speed of the actual gas turbine operating state under the corresponding steady state are collected, and the above variables are combined into a one-dimensional vector as a feature vector, and the feature vector is input into the SVM component degradation classification model of the low-pressure compressor, high-pressure compressor, combustion chamber, high-pressure turbine, low-pressure turbine and power turbine respectively to determine whether the low-pressure compressor, high-pressure compressor, combustion chamber, high-pressure turbine, low-pressure turbine and power turbine components are degraded. If it is determined that the performance of any of the above components is degraded, it is necessary to identify the degradation degree of the degraded component.

[0022] The identification process of the PSO algorithm is as follows: the degradation factor of the component with performance degradation is taken as the parameter to be solved, the initial population of degradation factors is randomly generated, and then the nonlinear mechanism model of the gas turbine is substituted to calculate the response characteristics of the inlet and outlet parameters of the gas turbine components corresponding to the corresponding initial population. The mean square error MSE of the calculated data and the measured inlet and outlet parameters of the gas turbine components is taken as the objective function, and the objective function value (fitness) is calculated. The formula is expressed as follows: , in: , , Indicates the operating status parameters of different components, corner marks Indicates measured data, subscript Represents the calculated data, Indicates the number of import and export parameters; Then, it is determined whether the objective function value reaches the termination condition. If so, the degradation degree is output. If not, the degradation factor of the population is updated based on the PSO algorithm, and a cyclic iterative calculation is performed until the set iteration number is met to complete the identification of the degradation degree of the degraded component.

[0023] In each iteration, each particle will determine its own direction and speed of action based on the optimal solution it has found (individual extreme value) and the optimal solution currently found by the entire population (global extreme value), update its own position, and then calculate the updated particle fitness until the set number of iterations is reached and the calculation is completed. Assume that in a d-dimensional search space, the population includes N particles, and the i-th particle is a d-dimensional vector, which can be expressed as , the movement speed of the i-th particle is also a d-dimensional vector, which can be expressed as , the individual extreme value of the i-th particle can be expressed as , the global extreme value of the entire particle swarm can be expressed as The update speed and position of the ith particle can be expressed as follows:

[0024]

[0025] In the formula, is the inertia weight; and is the acceleration constant; and is a random number in the range [0,1].

[0026] By introducing random factors in the iterative solution process, the global search capability of the algorithm is improved; in the iterative optimization process, as the iteration number increases, the inertia weight is adaptively adjusted. In the early stage of iteration, a smaller inertia weight is used to improve the global search capability of the algorithm. When it is close to the set iteration number, a larger inertia weight is used to improve the convergence of the algorithm.

[0027] Finally, if Figures 2 to 6 As shown ( Figure 2 is the classification accuracy of low-pressure compressor degradation, Figure 3 is the high pressure compressor degradation classification accuracy, Figure 4 is the high pressure turbine degradation classification accuracy, Figure 5 is the classification accuracy of low-pressure turbine degradation, Figure 6 is the power turbine degradation classification accuracy), the accuracy of the degraded component judgment obtained by the present invention is significantly higher than that of the two existing algorithms, Gaussian Bayes (GB) and random forest (RF), and the effect is more significant;

[0028] The above table is a table of identification errors when performance degradation occurs in different components under different experimental groups. It can be seen from the above table that the relative errors of each group of experimental data are relatively small, which shows that the accuracy of the identification results of the present invention is extremely high.

[0029] The SVM-PSO algorithm can accurately identify the states of up to three components that are degraded at the same time. By locating the degraded components, it reduces the number of performance degradation factors that need to be identified and the search dimension of the performance degradation factor identification space. It not only improves the accuracy of ship gas turbine performance degradation identification, but also improves the solution speed of the identification process.

[0030] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

[0031] This patent is not limited to the above-mentioned optimal implementation mode. Anyone can derive other forms of nighttime street scene semantic segmentation methods based on difficult category perception mechanisms under the inspiration of this patent. All equal changes and modifications made according to the scope of the patent application of the present invention should be covered by this patent.

Claims

1. A method for identifying performance degradation of a marine gas turbine based on SVM-PSO, characterized in that: Firstly, the inlet and outlet parameters of gas turbine components are collected, and then the components with performance degradation are determined through the SVM model. Finally, the PSO algorithm is used to identify the degradation degree of the degraded components, thereby realizing efficient and accurate identification of gas turbine performance degradation.

2. The method for identifying performance degradation of a marine gas turbine based on SVM-PSO according to claim 1, characterized in that: The process of establishing the SVM model is as follows: First, the Newton-Raphson algorithm is used to combine the principles of thermodynamics and kinetics to establish the nonlinear mechanism model of the gas turbine; then, the flow degradation factor and efficiency degradation factor are implanted in some component modules of the nonlinear mechanism model of the gas turbine, and the combustion efficiency degradation factor is implanted in other component modules. The range of the degradation factor is set to 0.95~1.0, and the response characteristics of the inlet and outlet parameters of the gas turbine components under different degradation factors are obtained. White noise is implanted in the obtained data, and labels are added to the data according to the set degradation factor. When the degradation factors of a certain component are all 1, it is in a non-degraded state and the label is set to 0. When the flow degradation factor or efficiency degradation factor of a certain component is less than 1, or the combustion efficiency degradation factor is less than 1, it is in a degraded state and the label is set to 1, thereby obtaining the SVM model data set; the SVM model data set is randomly divided, 80% is used as the training set and 20% is used as the data set, and the kernel function of the SVM model is set to a cubic polynomial kernel function, and the SVM model is trained, and corresponding SVM models are established for different components.

3. The method for identifying performance degradation of a marine gas turbine based on SVM-PSO according to claim 2 is characterized in that: The inlet and outlet parameters of the gas turbine components under steady-state operation are combined into a one-dimensional vector as a feature vector, and the feature vector is input into the SVM model to determine whether the component has been degraded. If it is determined that the performance of one of the above components has been degraded, the degree of degradation of the degraded component needs to be identified.

4. The method for identifying performance degradation of a marine gas turbine based on SVM-PSO according to claim 3 is characterized in that: Based on the collected inlet and outlet parameters of gas turbine components, the gas turbine mechanism model is verified and corrected.

5. The method for identifying performance degradation of a marine gas turbine based on SVM-PSO according to claim 2 is characterized in that: The identification process of the PSO algorithm is as follows: the degradation factor of the component with performance degradation is taken as the parameter to be solved, the initial population of degradation factors is randomly generated, and then the nonlinear mechanism model of the gas turbine is substituted to calculate the response characteristics of the inlet and outlet parameters of the gas turbine components corresponding to the corresponding initial population. The mean square error MSE of the calculated data and the measured inlet and outlet parameters of the gas turbine components is taken as the objective function, and the objective function value is calculated. The formula is expressed as follows: ， in: , , Indicates the operating status parameters of different components, corner marks Indicates measured data, subscript Represents the calculated data, Indicates the number of import and export parameters; Then, it is determined whether the objective function value reaches the termination condition. If so, the degradation degree is output. If not, the degradation factor of the population is updated based on the PSO algorithm, and a cyclic iterative calculation is performed until the set iteration number is met to complete the identification of the degradation degree of the degraded component.

6. The method for identifying performance degradation of a marine gas turbine based on SVM-PSO according to claim 5 is characterized in that: The PSO algorithm is optimized by introducing random factors in the iterative solution process to improve the global search ability of the algorithm. In the iterative optimization process, as the iteration number increases, the inertia weight is adaptively adjusted. In the early stage of the iteration, a smaller inertia weight is used to improve the global search ability of the algorithm. When it is close to the set iteration number, a larger inertia weight is used to improve the convergence of the algorithm.

7. A method for identifying performance degradation of a marine gas turbine based on SVM-PSO according to any one of claims 1 to 6, characterized in that: The inlet and outlet parameters of the gas turbine components include any one or more of the following: the low-pressure compressor outlet temperature, low-pressure compressor outlet pressure, low-pressure compressor flow, low-pressure compressor speed, high-pressure compressor outlet temperature, high-pressure compressor outlet pressure, high-pressure compressor flow, high-pressure compressor speed, fuel flow, power turbine outlet temperature, power turbine outlet flow and power turbine speed of the gas turbine.

Images