Research And System Implementation Of Solar Cells Photoelectric Conversion Efficiency Prediction Model Based On Machine Learning

Solar cell photoelectric conversion efficiency is an important parameter to evaluate the performance of solar cells, so accurate prediction of PCE has great significance to improving the performance of dye sensitized solar cell. However, due to the complexity of the structure of the battery, it is difficult to get the experimental value of the photoelectric conversion efficiency from the molecular structure mechanism through the quantum chemistry method. The intelligent calculation method can avoid the complex reaction process of quantum chemistry method. Therefore calculation of the physical and chemical properties of sensitized dye molecules using quantum chemical method, then based on the intelligent calculation method to establish the regression model between the molecular physical and chemical properties, structure properties and photoelectric conversion efficiency, and the predicted value of the photoelectric conversion efficiency is obtained.So the main work of this paper is to calculate the physical and chemical properties of the ground state and excited state of the dye molecules on the STO-3G and 6-31G(d) basis using quantum chemical B3LYP method. Then use the Support Vector Machine with five kinds of feature selection methods(Simple Linear Regression, Sequential Forward Selection, Sequential Backward Selection,+N-n algorithm, Mean Impact Value) to establish the regression model between the molecular physical and chemical properties, structure properties and the photoelectric conversion efficiency. In this paper, Jsc, Voc and FF as a bridge to connect the relationship between molecular descriptors and PCE, and is used to establish the cascade regression model. Through the good evaluation criteria of the model as the theoretical support, based on B/S mode using the JSP language developed photoelectric conversion efficiency of solar cells prediction system. This prediction system can accurately predict the PCE, and then modify the molecular structure, to further improve the efficiency of the dye molecules, and finally can be designed to synthesis and device research.