Quantitative fund transaction research based on fractional order neural network and Deep-Q Network

摘要

Quantitative fund trading strategies can contribute to investors by reducing the risk more effectively and maximizing their own interests. Fractional order neural network and Deep-Q Learning network can be used to make trading decisions in this paper. Firstly, in order to make Deep-Q Learning network converge faster, a new data preprocessing method is proposed. Fractional order neural network is trained by the preprocessed data, and the improved gradient update method can be used to update the network parameters. Then the problem that negative numbers have no fractional power can be solved by Separation of Symbols. Finally, the trained Deep-Q Learning network can make trading decisions and get some returns. The transaction results show that the proposed method is effective and stronger than the traditional method.