Font Size: a A A

Research On The Construct Of LSTM-RNN International Crude Oil Price Prediction Model Based On Influence Characteristics

Posted on:2021-02-15Degree:MasterType:Thesis
Country:ChinaCandidate:Y T ZhuFull Text:PDF
GTID:2370330623483849Subject:Management Science and Engineering
Abstract/Summary:PDF Full Text Request
Crude oil is one of the main driving forces of economic development.In the prediction of crude oil prices,the linear method represented by econometrics has been greatly developed.With the rise of machine learning,we can make predictions with nonlinear methods especially SVM and neural network.However,in the application of these methods,they either focus on the self-limiting factors such as financial or its energy nature,or ignore the endogenous influence characteristics of each factor,or they are too single in the model type selection.This paper,based on the dual attributes of crude oil,constructed the TVP-VAR model to determine the impulse response of each factor to oil price,clarify the influence characteristics of each factor on oil price,and provide a basis for index selection and parameter setting of LSTM-RNN model.By constructing LSTM,RNN and control models,the LSTM-RNN international crude oil price prediction model based on the influence characteristics is obtained.In this study,by comparing and analyzing other prediction methods,the model we contracture is more superior.Conclusions are as follows:(1)commodity attribute factors and financial attribute factors will affect the price of crude oil and have time-varying effects.But the effect of coal price,CPI and RMB exchange rate on crude oil price is not significant.The impact of various factors on oil price is most significant when it lags behind for 3 month,and it converges when it lags behind for 6 months.(2)compared with traditional econometrics,machine learning methods has better performance.LSTM-RNN have smaller errors compared with BP neural network model and SVM and model.(3)the LSTM model performs better in predicting the crude oil futures price;and RNN model performs better in predicting spot oil prices.On the basis of the above research,some suggestions and future works are put forward,such as promote the influence of RMB in the international market,pay attention to the financial property of crude oil,and improve the construction of crude oil database.
Keywords/Search Tags:Prediction Model of Crude Oil Price, Double Attribute of Oil, TVP-VAR, LSTM-RNN
PDF Full Text Request
Related items