With large-scale access to distributed photovoltaics, distribution network bus load forecasting gradually transforms into net load forecasting. To reduce the impact of distributed photovoltaics' difficulty in obtaining accurate meteorological data on the accuracy of net load forecasting, a method is proposed to separately predict distributed photovoltaic output and load, and use the hierarchical relationship between net load, load, and photovoltaic output to correct the prediction results. net load forecasting method. First, predict the output of centralized photovoltaic power stations that have a high correlation with distributed photovoltaic output and have relatively complete meteorological data, then calculate the optimal conversion coefficient between the two, and finally obtain the prediction results of distributed photovoltaics through conversion. For load prediction, the Attention Bidirectional Long Short-Term Memory (Attention-BiLSTM) prediction model is used. After obtaining the distributed photovoltaic and load prediction results, the hierarchical correction algorithm is used to correct the photovoltaic and load prediction results. Using real data in a certain area as an actual calculation example, the experimental results show that the model proposed in this paper is more effective in net load prediction of distribution networks that lack accurate meteorological data.