Current common bus load forecasting methods use statistical modeling of historical data, ignoring many influencing factors and differences in load variation characteristics, resulting in large forecasting errors and poor timing. To optimize the above problems, a medium and high voltage bus load forecasting algorithm considering structured electricity consumption interactions is investigated. The user-side characteristic indicators of medium- and high-voltage bus load are selected with the structured interaction information generated by customers when they use electricity. After initial processing of load data, the data are decomposed to avoid interference from internal factors. A genetic algorithm improved RBF network is used to implement the prediction of medium and high voltage bus loads. The algorithm test results show that the studied algorithm error is reduced by about 1.46% on average, and the efficiency is significantly improved with good prediction results.