The Short-term And Long-term Forecasting Of Methanol Price Based On Phase Space Reconstruction And Extreme Learning Machine

As an important and basic organic chemical raw material,methanol is used in the national economy widely,the fluctuation of its price influences on the other chemical products on market significantly.However,the price of methanol is affected by many factors,how to improve accuracy of the short-term and long-term price prediction has been focused by researches widely.In this paper,firstly several more noteworthy elements are selected from all the elements which can influence the price of methanol by analyzing the actual market data.Secondly the feature information of elements that abstracted by phase space reconstruction was used to predict the price in short-term and long-term based on Extreme Learning Machine(ELM)with fast learning speed and global approximation performance.Effectiveness of the algorithm was proved by the computer simulation,which can provide a theoretical guidance for predicting the actual price of methanol on market.The first innovation is that the phase space reconstruction was used to abstract the important information between all the data based on the actual data that which elements influence methanol price.The second innovation is that a neural network learning algorithm with fast learning speed and global approximation performance-extreme learning machine(ELM,Extreme Learning Machine)is used to predict the methanol price of the short-term and long-term.Effectiveness of the algorithm was proved by the computer simulation.The first chapter of this paper is introduction,the price of methanol,research background and situation were introduced at this chapter.The second chapter is time series prediction theory.The feature and classification of time series,moreover,the parameters of delay time and embedding dimension in phase space reconstruction were introduced at this chapter.The third chapter is prediction theory of artificial neural network.In this chapter,firstly,the history and basic characteristic of artificial neural network were introduced briefly.Then,different kinds of algorithms of artificial neural network were introduced.Finally,the extreme learning machine and the dependency problem of its activation function were introduced emphatically.In fourth chapter,firstly,the price,import volume,import average price,export volume,export average price of coal were introduced as time series variable,the correlation analysis between the variables above mentioned and price of methanol.Secondly,the embedding dimension and delay time parameters of each time sequence are determined using phase space reconstruction.With the corresponding input data,a neural network model based on extreme learning machine is established to predict the price of methanol.Chapter 5 combines the advantages of particle swarm optimization(PSO)and gravity search algorithms(GSA)to optimize the weight and bias of extreme learning machine algorithms.The prediction accuracy of methanol prices is further improved by means of the PSO-GSA-ELM algorithm.The computer experiments based on MATLAB in chapter 4 and chapter 5 show that the methanol price prediction model established in this paper has good prediction accuracy and good data tracking ability,and can be applied to practical methanol price forecasting occasions.