除了星系的光谱红移之外， 星系测光红移的估计也对研究宇宙大尺度结构及演变有着重要的研究意义。 利用斯隆巡天项目最新发布的SDSS DR13的150 000个星系的测光及光谱数据， 在红移值Z<0.8范围内， 先使用SOM自组织神经网络对星系样本进行早型星系和晚型星系的聚类， 然后用遗传算法优化后的BP神经网络对星系的测光红移进行估算。 估算结果与作为标准的已知星系光谱红移进行比对， 早型星系的红移估计最小均方误差约为0.001 3， 晚型星系最小均方误差约为0.001 7。 实验结果表明， 遗传优化的BP算法在精度上优于BP神经网络算法， 且效率上优于K近邻、 核回归等传统测光红移估计算法。

In addition to the spectral redshift of galaxies, the photometric redshift estimation of galaxies has important implications for the study of large-scale structures and evolution of the universe. In this paper, it chose about 150 000 galaxies’ photometric and spectral data in the latest SDSS DR13 of the Sloan survey project within the spectral redshift range of Z<0.8. The SOM self organizing neural networks were used to cluster galaxies in early type galaxies and late type galaxies. And then the photometric redshift of the galaxies was predicted by the BP neural network optimized by genetic algorithm. The prediction results were compared with the spectral redshift of galaxies. The mean square error of the redshift estimation of early type galaxies was about 0.001 3, and it for the late type galaxies was about 0.001 7. Experimental results showed that the BP algorithm optimized by genetic algorithm was more accurate than the BP neural network algorithm, and was more efficient than K nearest neighbor and kernel regression algorithms.