Research On Coal Price Forecasting Models Based On LSTM

As a core product of related companies,coal price prediction plays an important role in corporate financial accounting.At the same time,accounting prediction is an important means of enterprise economic management.Its purpose is to quantitatively or qualitatively judge,speculate and plan the development and changes of internal economic activities of enterprises.Laws and evaluations to guide and regulate economic activities and seek the best economic results.Therefore,the study of coal prices has a profound impact on the financial management of related companies.In recent years,experts,scholars for the study to predict the price of coal is mostly to regression fitting and econometrics etc.The price of the traditional prediction method is given priority to,although some scholars introduce the method of artificial intelligence price prediction field,but also based on BP neural network,expand research is not in-depth study LSTM model in the application of the coal price forecast.The coal price prediction model based on LSTM neural network can well fit the short-term coal price trend in China,and then provide a certain basis for decision makers.In order to build a prediction model that is in line with the mathematical statistical characteristics of coal price in China,this paper takes qinhuangdao thermal coal(Q5500)weekly market price as the research object and conducts in-depth research from the following four aspects:(1)In-depth analysis of the influencing factors of coal prices in my country shows that the correlation between various variables is high and it is not easy to quantify and fit.Therefore,this paper chooses the single variable of coal price for model construction;(2)The intrinsic mathematical characteristics of coal price are studied,and it is concluded that its traditional prediction model with nonlinear characteristics has a poor fitting effect,so this paper chooses the LSTM neural network model as the basic model;(3)In-depth analysis of the principles and ideas of the LSTM model,it is concluded that LSTM will be as easy to fall into the problem of local optimal solution and slower convergence speed as other neural networks;(4)Compared with the previous people,optimizing the LSTM model only from the learning rate cannot completely solve the problem.This article starts from the model input terminal and the model optimization function to improve it.After empirical testing:using the Lookahead optimization algorithm to optimize the LSTM model,without reducing the model's efficiency,the model's convergence speed is greatly improved;Based on the improvement of the data input terminal,wavelet transform and Kalman filtering were used for fitting,respectively,to reduce the prediction error from the basic 3.28 yuan/ton to 1.55 yuan/ton and 3.04 yuan/ton.The above research shows that the coal price prediction method of the improved LSTM series model has good practicability and prediction accuracy.While removing the coal price noise data,it can well retain the trend and periodicity of the price data itself.The problem of coal price prediction under the background of big data provides new ideas.