The bus load is highly volatile and susceptible by the influence of users" electricity consumption behavior. Affected by the Distribution Generator (DG), the uncertainty of net load of the bus will further increase. In order to solve this problem, a new method of net bus load prediction using Random Forest (RF) as a predictor to predict photovoltaic DG output and bus load respectively is proposed. Firstly, the high-dimensional original feature set including meteorological, social information and other factors is constructed. RF predictors of photovoltaic DG output and bus load are constructed respectively based on the original feature set. In the training process of RF, the importance of each feature in original feature set was analyzed and sorted by PI value. Secondly, RF model prediction accuracy of characteristic subset of different dimensions is taken as the decision variable,