In order to improve the communication quality of the broadband power line communication system and based on the basic principle of the broadband power line communication system, a simulation model of the broadband power line communication system is built. Taking the actual data collected from a district user’s electric meter in Yunfu of Guangdong Province as the original data, the BP neural network and Elman neural network are introduced to the simulation test of the communication quality under the 500 m four-path channel model. Aiming at the disadvantage of poor anti-noise performance of neural network algorithms, an Elman neural network based on improved VPGA optimization is proposed for the data processing of the de-mapping module of power line communication system, and the simulation test is carried out. Experimental results show that the algorithm does not occupy valuable spectrum resources and is convenient to implement. Except the extreme bad channel environment, the algorithm can significantly improve the communication quality of the broadband power line communication system and reduce the bit error rate. This work is supported by Science and Technology Project of China Southern Power Grid Company (No. 035300KK52150007).