One of the key steps for optimal bidding in power markets is to estimate the rivals’ bidding behaviors. However, for most participants, it would be difficult to directly forecast the rivals’ individual bids due to the information privacy and volatile characteristics of individual bidding behaviors. From another point of view, the aggregation of individual bids, denoted as aggregated supply curve (ASC), might be helpful to offset the uncertainties of individual bidding behaviors and can be used as reference for optimal bidding. In fact, the real ASC data contains bidding information from thousands of participants, which would be formulated with high dimensionality and unstructured formats, not applicable for general forecasting methods. Thus, a novel data-driven ASC forecasting framework based on long-short term memory (LSTM) model and corresponding data processing techniques is proposed in this paper. In detail, A paradigmatic data integration method is proposed to fix the unstructured data formats. A feature extraction method is developed to simplify the high dimensionality of ASC. Then, a LSTM model is customized to forecast ASCs. At last, real data from Midcontinent Independent System Operator market in the U.S. are utilized to demonstrate the forecasting performance of the proposed framework.