Abstract: In recent years, the Internet financial market has achieved rapid development across the globe. In the meantime, Internet finance has become a hot topic in academia. Compared with traditional financial markets, the Internet financial market has higher liquidity and volatility. In this paper, the dynamics (daily trading amount and count) of the Internet financial market is studied and a prediction model is proposed based on deep neural network for fusion hierarchical time series learning. Firstly, the model can process the multiple sequence (macro dynamic sequence and multiple subsequences) feature as the input variables. And then, an attention mechanism is proposed to fuse the input variables from both the time and subsequence feature dimensions. Next, the model designs an optimization function based on the stability constraint of the sequence prediction, which makes the model have better robustness. Finally, a large number of experiments have been carried out on real large-scale data sets, and the results have fully proved the effectiveness and robustness of the proposed model in the dynamic prediction of Internet finance market.