Abstract: Aimed at the problem that traditional trading strategies cannot obtain stable returns under complex market conditions, a DQN risk control trading strategy based on compensation standard deviation is proposed. By fusing historical market data and technical indicator data, using convolutional neural network to extract data features, judging transaction signals, and using the cumulative compensation standard deviation to calculate the reward function with risk regulation, we effectively improved the adaptive ability of the strategy. This strategy conducted trading experiments on the Shanghai and Shenzhen 300 Index from 2015 to 2019. In the 2019 test phase, the strategy's annual return reached 16.13%, the winning rate was 54.62%, and the Sharpe ratio was 15.91%.