Analysis and Construction of EEMD Smart Model and Fuzzy Forecasting through Improved Bayesian Estimation
Liyao Tang, Zijian Cui, Kaishuo Liu · Journal of Physics Conference Series · 2021
Abstract Different portfolio models will have a great impact on the model rate of return. In order to meet the data needs of portfolio investment forecasting, a big data generation method based on incremental Bayesian network model is proposed. Firstly, the fuzzy interval number is constructed by using the global GM (1pare 1) prediction model, then the fuzzy Mmurv model is established based on the interval fuzzy number, and the model is optimized based on the midpoint, radius and acceptability of the interval number, thus a single objective programming model with parameters is obtained. The future data is partially generated by using the time series generation algorithm, and the Bayesian network model trained by historical data is updated with the newly generated data. So that the updated Bayesian network can reflect the relationship between the variables and the laws contained in the new and old financial data in this period of time. And use the model to predict the amount of foreign direct investment absorbed by China in 2020, test the robustness and analyze the impact of major random events to show the reliability of the prediction. The analysis is carried out on multiple scales. The results show that the integration of IMF1 and IMF2 components through the white noise test can be regarded as the random impact of external factors; the cycle length of IMF3 component is about 3 years, which can be regarded as the inventory cycle caused by the change of enterprise inventory; the cycle length of IMF4 component is 5 years, which can be regarded as political cycle; the cycle length of IMF5 component is about 20 years, which can be regarded as construction cycle. The set of portfolio paths is generated by path search algorithm in Bayesian network, and the big data set with real data characteristics is generated according to the probability distribution of each path. The experimental results show that the method is feasible and ensures a certain accuracy. The forecast results show that the amount of foreign direct investment in China will continue to fluctuate seasonally and in a trend in recent years.