Temperature Field Reconstruction Based On Dimension Reduction And Deep Learning Method

In current time,most of China's power supply comes from fossil fuel combustion.With China's new target of reaching peak carbon dioxide emissions before 2030 and achieving carbon neutrality before 2060,the thermal power industry is facing a formidable challenge.Tapping the potential of energy-saving and emission-reducing in the combustion process is an effective measure for clean utilization of fossil fuel.It is also the only path for thermal power industry to embrace green development.In combustion process,temperature shows the state of energy transformation and' energy transmission.Obtaining the temperature field information rapidly and accurately plays an important role in optimizing combustion process,improving combustion efficiency and reducing pollutant emission.For this,research on temperature field measurement has a significant meaning in combustion process.For the purpose of developing a new method,that can use a few sensors to get a rapid and reliable reconstruction of the temperature field.The main work of this research can be summarized as follows:(1)Based on data dimension reduction algorithms and deep learning algorithms,new temperature field reconstruction methods are proposed in this research.As a basis of this research,the advantages and disadvantages of different methods are compared.Then,linking the existing research experience and temperature measurement needs,data dimension reduction algorithms and deep learning algorithms are introduced into the research of temperature distribution reconstruction.New reconstruction methods combining the advantages of computing methods and measurement methods are proposed.After that,numerical experiments and methane combustion simulation model are set up to test the feasibility of proposed methods.Test results show that both of the proposed methods can result in a rapid and reliable reconstruction solution and can be used in temperature measurement.(2)Optimization methods for the key parameters are proposed.Since the validity of the proposed methods are determined,to improve the accuracy and stability of proposed methods,the key parameters of reconstruction calculation are optimized by simulation data,such as the numbers of eigenvectors,core tensor dimensions.The relationship between the key parameters and the accuracy is determined.Furthermore,optimization methods are proposed.After optimization,the accuracy and the stability of the proposed methods are noticeably improved.(3)For solving the problem of using inaccurate prior data,the calculation process of the proposed methods is improved.First,the significance of using inaccurate prior data to get reconstruction solutions is analyzed.Then,the proposed methods are improved by optimizing the calculation procedures.After that,the experiment data is used in the calculation process.The differences between reconstruction result and measurement data are compared.Although the reconstruction errors are larger than previous test results,the consistency of temperature field between reconstruction result and the measurement data shows that capability of the proposed methods in the experiment temperature field reconstruction.It is also proved that proposed methods can be used in real application.In conclusion,based on data dimension reduction algorithms and deep learning algorithms,for solving temperature field reconstruction problem,new reconstruction methods are proposed in this study.The objective of developing a new method that can use a limited number of sensors to get a rapid and reliable reconstruction of the temperature field is realized.It also provides a new method for further temperature measurement research.