Research on Digital Financial Sequence Analysis and Forecasting Based on LSTM Algorithm Optimization

Current methods for digital financial time series analysis and prediction have defects such as low efficiency and large error, so the study proposes an optimization model for digital financial time series analysis and prediction by integrating the whale optimization algorithm, seasonal trend decomposition, and long and short-term memory network for this kind of problems, and verifies its effectiveness. The experimental results show that the average absolute error value of the research model is 0.013 and the value of the coefficient of determination is 0.975 in the comparison between the research model and other models, which are better than the comparison model. In the comparison of trend prediction of different models, the trend fitting effect of the comparison model a is better, but the predicted value has a large fluctuation; model b has a better nonlinear ability, but does not have the ability of long-term memory, so it is difficult to capture and learn the potential information in a long time series. Model c with the addition of Baidu index has a lower prediction error, but is significantly worse than the research model. In the actual RMB exchange rate prediction, the average absolute error value of the research model is 0.001 and the root mean square error value is 0.002, which is better than the comparison model. Taken together, the research model possesses effectiveness in analyzing and forecasting the time series of RMB exchange rate in digital finance, which can effectively enhance the actual digital finance time series forecasting ability.