Natural Gas Price Predicting Adopting CNN-LSTM Hybrid Neural Network Model

Abstract

Natural gas price affects a lot in people’s daily expenses. Sudden change in natural gas influences people’s life to some extent. The study experiments and compares the CNN, LSTM, and CNN-LSTM hybrid model’s performance, and applies them to predict and forecast natural gas prices. CNN-LSTM model increases the number of neurons filters and use denser and deeper network structures. From the result, CNN-LSTM showed a significantly better reduction in average loss than a single model. The hybrid model also fitted a closer curve to the natural gas price line. The multivariate hybrid model gives more accurate result than the single model. The prediction method used in this paper can give better insight into evaluating natural gas price trends and provide people with ahead forecast of prices. The paper offers a basic reference to the direction of future improvements of model structures in further study.