Computer Intelligent Investment Strategy Based on Deep Reinforcement Learning and Multi-Layer LSTM Network

Investors frequently make business about volatile assets with the goal of maximizing their profits. This aim of this article is to establishes a model to study the investment strategy of bitcoin and gold. It's not always accurate to use algorithm based on traditional time series model to predict the prices. Therefore, The multi-layer LSTM weighted model of multi-layer LSTM network this article puts forward is used to study the input time span to factor the influence of output time span. By using the embed approach, this model can achieve higher accuracy compared with LSTM model by influencing the stacking sequence generation cycle and training model. Meanwhile, this model introduced the Attention mechanism and can make short-term and long-term forecasts continuously. It also quantified the fluctuation of MACD, RSI, Apriori and other parameters to construct the risk measurement index. The higher the periodic price fluctuation is, the greater the risk can be. The results show that timely business maximizes profits. An initial $1,000 investment on September 10, 2021 can even profit more than $900,000. In order to prove the model provides the best strategy, depth of reinforcement learning model is set up. Bitcoin and gold history feature information of state data matrix and agent within the depth of neural network is utilized to extract the stable characteristics of data expression. Based on these features, this model utilized optimized strategy method of three reinforcement learning algorithms. Combined with the current reward, it optimized the relevant parameters of the deep neural network to obtain the optimal investment strategy in exploratory ways. Moreover, the model will be ran twice to test the sensitivity under different transaction costs (investment decision results) by setting points of 0.5%, 1%, 1.5% and 2% according to the percentage of increase and decrease of transaction costs. The results show that the model has good robustness, which means the model constructed in this paper has good results under different transaction costs.