Daytime Radiative Cooling Structural Design By Machine Learning And Genetic Algorithm

With the rapid development of society and the gradual improvement of people’s living standards,a series of environmental problems have emerged one after another.The massive emissions of greenhouse gases have led to global warming,and the widespread use of air conditioning equipment has led to an urgent need to address the issue of electricity consumption.Radiation refrigeration is a new type of refrigeration technology that utilizes the physical mechanism of radiative heat transfer to achieve low-temperature cooling.Its biggest feature is that it can achieve cooling effects without any additional energy input,and the cooling factor in the radiation refrigeration system is a natural medium without any harm to the environment.Therefore,it is considered the most potential alternative refrigeration technology for traditional refrigeration.Radiation refrigeration does not require the commonly used refrigerants and mechanical compression equipment in traditional refrigeration technology,and has the advantages of no pollution,low noise,and high efficiency.The key to radiation refrigeration technology lies in the selection and preparation of radiation materials,which can be made of metal or semiconductor materials,and optimized for their optical and thermal properties through structural design such as nanostructures and multi-layer film structures.At present,radiation cooling technology has been widely applied in fields such as building cooling,photovoltaic cell cooling,and personal thermal management,and its application prospects are even broader.In this article,machine learning methods（MLM）and genetic algorithms（GA）were used to design and optimize a daytime radiation refrigeration transmitter composed of polydimethylsiloxane（PDMS）,silicon dioxide（Si O₂）,and aluminum nitride（Al N）,and the effects of different factors on the refrigeration performance of the radiation cooler were explored.The main research content and results are as follows:（1）In this article,we combine machine learning methods with genetic algorithms to calculate and optimize the design of a daytime radiation cooler structure,which covers PDMS,Si O₂,and Al N from top to bottom on an Ag Si substrate to achieve daytime radiation cooling.The spectral performance and refrigeration performance of the designed daytime radiation cooler were accurately calculated and comprehensively analyzed using the transfer matrix algorithm and FDTD Solution.（2）The material data set and machine learning model suitable for this work were selected through feature engineering and machine learning model evaluation.Based on the"experience"obtained through machine learning training,the spectral performance of the optimal daytime radiation cooler can be accurately predicted,with an average prediction error of only 0.56%.In addition,using machine learning combined with genetic algorithm optimization to obtain the optimal radiation cooler has a calculation time 67 times faster than the transfer matrix algorithm,and has extremely high computational efficiency while ensuring high machine learning prediction accuracy.（3）The spectral performance of the optimal daytime radiant cooler designed in this paper is calculated by the transfer matrix method.The modified cooler structure has an average total spectral hemispherical emissivity of 94.43%in the atmospheric window band and an average total spectral hemispherical reflectance of 98.25%in the solar radiation band.The reason for this spectral performance is that the three selected emission layer materials all have large refractive index real part and imaginary part,and they form complementarities between the high emission regions of the atmospheric window band,which greatly increases the wide band emissivity of the cooler structure in the atmospheric window band.When the ambient temperature is 30℃and the solar irradiation power is 900 W/m²,the net cooling power of the optimal daytime radiation cooler can reach 140.38 W/m².In theory,the daytime temperature of the cooler is 9.08℃lower than the surrounding ambient temperature.This article applies machine learning methods to the theory of radiation refrigeration and designs daytime radiation refrigerators with strong refrigeration performance.Compared to traditional spectral performance calculation methods for refrigerators,machine learning methods have faster calculation speed while ensuring high prediction accuracy.Most importantly,this article provides an important research example for the application of machine learning in the design of radiation coolers.