An Online-Calibrated Time Series Based Model for Day-Ahead Natural Gas Demand Forecasting

Hadi Khani, Hany E.Z. Farag

    Research output: Contribution to journalArticlepeer-review

    17 Scopus citations


    This paper proposes a new online-calibrated time series based model with the application to the day-ahead natural gas demand (GD) forecasting. A double-stage parallel process is developed for creating the forecasting model. The two stages include analysis of the temperature-independent and temperature-dependent components of the GD. The former stage is executed by online processing of the historical GD information considering the intertemporal variation of the GD. The latter stage, however, is conducted by exploiting the features of the GD information correlated with the ambient temperature. The forecast of the temperature is incorporated into the GD forecasting model through a correlation-based function. The model can generate the day-ahead GD forecast with both the hourly and intrahourly resolutions without compromising the forecast accuracy. The model is calibrated online using the historical GD and temperature information to achieve a higher forecast accuracy. The practical challenges associated with the industrial application of the model are also discussed. The application of the proposed model is numerically examined using real-world GD and temperature data, and the results are comprehensively studied. The outcomes reveal the efficacy and feasibility of the proposed model under various cases.

    Original languageBritish English
    Article number8423503
    Pages (from-to)2112-2123
    Number of pages12
    JournalIEEE Transactions on Industrial Informatics
    Issue number4
    StatePublished - Apr 2019


    • Correlation-based functions
    • day-ahead forecasting
    • gas demand (GD)
    • industrial applications
    • information process
    • online calibration
    • parallel process
    • time series based functions


    Dive into the research topics of 'An Online-Calibrated Time Series Based Model for Day-Ahead Natural Gas Demand Forecasting'. Together they form a unique fingerprint.

    Cite this