Daily natural gas price forecasting by a weighted hybrid data-driven model

With the role of natural gas gaining increasing importance in the transition of the world energy system and addressing global climate change, accurate prediction of the price of natural gas becomes crucially important. This paper first introduces three widely used individual data-driven models, i.e., support vector regression (SVR) and long-term and short-term memory network (LSTM), and a modified data-driven model, i.e., the improved pattern sequence similarity search (IPSS). A new weighted hybrid data-driven model based on these three models is then proposed. To train the model, data regarding the daily natural gas spot price in the U.S. prior to June 2018 are used and the model's prediction ability is tested using data from June 2018 to May 2019. The results show that the new IPSS model can predict the daily price of natural gas accurately. In a comparison of prediction errors with other individual models, the proposed hybrid model demonstrated the highest prediction ability of all of the investigated models. •A novel hybrid model combining IPSS, SVR and LSTM is proposed.•The variance reciprocal method is used to determine the weight of the hybrid model.•The US daily spot gas price is used to test the model and the results are compared.•Comparison results indicate that the hybrid model has the highest forecast ability.