Aiming at the defect that traditional channel estimation technology will occupy spectrum resources, an improved algorithm based on improved random forest is proposed. Firstly, the improved SMOTE algorithm is preprocessed according to the inherent imbalance of the source data. Based on the characteristics of the power line channel, the synthetic rules of the minority data are determined, and the performance of the subinterval error rate of the mapped module is analyzed as the evaluation index. A broadband power line communication system model is built, and the actual meter data is used as the source data to carry out the simulation test under 500 m 18 path power line channel model. The experimental results show that the algorithm can make up for the influence of the inherent imbalance of the meter data on the performance of the random forest, and greatly reduce the fluctuation of the bit error rate. In all kinds of signal to noise environment, the improved random forest algorithm can well optimize the performance of the module, improve the quality of broadband power line communication and reduce the bit error rate. This work is supported by Science and Technology Project of China Southern Power Grid (No. 035300KK52150007).