Finance is not only the lifeblood of an economy, but also the lever to adjust the macro-economy. A modern economy is a market economy and essentially a developed financial economy. Based on the analysis of the problems faced by traditional finance and the overview of smart finance, this study puts forward the application of deep learning combined with reinforcement learning in smart finance to solve the problems existing in financial activities for the first time, and verifies through experiments. The model has better data and information processing ability compared with the traditional financial analysis mode. It provides higher quality decision-making information and bring more benefits. Taking a bond rating report as an example, it usually takes about 2 hours for manual in-depth analysis and carding, while it only takes about 2 minutes to interpret and refine the report by using the deep reinforcement learning model. The model has a certain reference value to solve the problems of traditional finance.