The nighttime light (NTL) data has the unique ability to detect the night light on the earth surface, which provides an indispensable data basis for monitor urban development and characterize the human activities. However, two of the most widely utilized NTL data sources, the Defense Meteorological Satellite Program Operational Line Scanning System (DMSP/OLS) from 1992 to 2013 and the Suomi National Polar-orbiting Partnership-Visible Infrared Imaging Radiometer Suite (NPP/VIIRS) from 2012 to now, were not consistent. To address this problem, we proposed a novel transformer-based DMSP/OLS super-resolution model, namely MHAT, to convert DMSP/OLS data into NPP-like data. It consisted of three modules: pixel embedding, feature extraction, and high-resolution image reconstruction modules. Specifically, we utilized the high-order attention module for deeper information mining on the DMSP/OLS data. The results showed that: (1) Our MHAT model achieved a PSNR of 52.40 dB and an SSIM of 0.97, demonstrating high precision. (2) The DN value distribution histogram of the super-resolution results was very consistent with that of the high-resolution NPP/VIIRS image, which proving the reliability of our MHAT model. (3) We took China as an example and calculated the correlation of SumDN derived from these two images for 385 cities in China. We found that there was a strong correlation between the super-resolution results and the high-resolution NPP/VIIRS image on the city scale (R2 = 0.94).