Abstract: This study takes the harmonic control project of a power grid in an industrial park in Shandong as an example, and explores in depth a harmonic source identification and control technology based on deep learning algorithms. After engineering verification, it was found that this scheme has a significant improvement effect on the precise positioning, dynamic identification, and rapid suppression of harmonic sources. It can achieve an automatic identification accuracy of 95.6% for harmonic sources, reduce the total harmonic distortion (THD) of voltage to 2.1%, and shorten the response time to 100 milliseconds. It has certain promotional significance for improving the power quality and grid operation reliability of industrial parks.