Natural Gas Price Forecasting using Statistical Models and Deep Learning Models

Abstract

Natural gas, entitled as methane gas or natural methane gas, is a highly flammable, colorless, odorless gaseous hydrocarbon where ethane and methane form the core. A petroleum resource which is associated with crude oil, burning it results in less emission of carbon which promotes a sustainable environment. In order to be extremely safe, go eco green, reduce dependency on nations for fuel resources it is even more tactical to forecast the prices of natural gases in the international market for a time frame. The present models shine either on statistical or deep learning models making it a. Thus, an ARIMA model is developed initially by using the updated dataset(Nasdaq) for the forecast to predict the closing price of the day. The autoregression predicts the upcoming values(closing price) based on then values. Moving Averages play a crucial role in smoothing the time series data. Secondly, the LSTM model is constructed with the same data frame. LSTM uses recurrent neural networks (RNN). The ideology behind the model is that at times being conscious of recent information to perform the present task. Bidirectional LSTM is also constructed. The Neural Prophet which is built on the top of pytorch is also experimented by means of forecasting. Neural prophet is extensively used by developers for the extension of the framework. The experimental repercussion showed that the proposed models are more efficient in terms of prediction and accuracy of the closing price