Session: 04-16 Combustor Design IV

Paper Number: 154311

Optimization of Combustor Based on Machine Learning Adaptive Dimensionality Reduction Method 

Gas turbine is an important power equipment in modern industry, and the combustion chamber is its key component. The geometric structure has a significant impact on its performance. Excellent geometric design can improve fuel combustion efficiency, reduce pollutant emissions, and maintain uniform outlet temperature, thereby extending the service life of turbine components. This article studies the typical geometric structure of combustion chambers and uses machine learning to find the optimal geometric parameters to improve performance.

 

The study used a real tubular combustion chamber. Firstly, a three-dimensional model was established using UG software. Then, geometric processing was performed using Ansys SpaceClaim, and Fluent Meshing was used to partition the mesh and simulate it. Parameters such as outlet temperature and combustion efficiency were tested to establish a high-precision combustion model.

 

To study the influence of multiple parameters, a full process simulation parameterization platform based on Ansys Workbench is built, and its advantages in full process, scalability, and universality are analyzed. Select 10 sets of parameters with the radial distribution coefficient of outlet temperature as the optimization objective, and generate 150 sets of test data using Latin hypercube sampling.

 

Build a four layer neural network model, optimize the combination of hidden layer nodes using Grid Search, calculate the R ² mean value through K-fold cross validation to obtain the optimal node combination, and analyze the influence of each feature parameter using SHAP to determine that the top 5 important feature parameters account for 85.12% of the total. To ensure the accuracy of the results, generate more data for calculation. Subsequently, the particle swarm algorithm was used to optimize in multidimensional space, reducing the radial distribution coefficient of outlet temperature by 30.5% to 0.107, and obtaining corresponding geometric structural parameters to achieve the optimization goal of the combustion chamber.

Presenting Author: Ao Sun Zhejiang University

Presenting Author Biography: PhD Candidate, majoring in combustor optimization and machine learning

Authors:

Ao Sun Zhejiang University
Kai Zhou Zhejiang University
Jianping Yan Zhejiang University