Intelligent forecasting of distribution system loads

Bahman Kermanshahi*, Ryuichi Yokoyama, Kazuhiro Takashashi

*Corresponding author for this work

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    6 Citations (Scopus)


    In this paper, for the first time, two algorithmic and one nonalgorithmic models have been developed for forecasting of distribution systems/feeders loads. Two algorithmic models are a 3 dimensional model of daily peak load versus daily peak wind and temperature and the a load versus wind-chill which has the advantage of reducing the analysis from a 3 dimensional model to a 2 dimensional model. The main criterion for the load forecasting study is that the final method used must have an average error of 5% or a curve fit above 0.9. It should be noted that 5% error is acceptable for mid-term load forecasting due to the accuracy of long-term weather forecast. The nonalgorithmic method is the application of neural networks. The reliability of the forecasts using the neural nets, combined with their ability to perform at this level without the aid of an experienced system operator, make neural nets an attractive alternative for load forecasting. Therefore, it has been selected for practical implementation in a power utility.

    Original languageEnglish
    Title of host publicationProceedings of the Mediterranean Electrotechnical Conference - MELECON
    EditorsM. De Sario, B. Maione, P. Pugliese, M. Savino
    Place of PublicationPiscataway, NJ, United States
    Number of pages4
    Publication statusPublished - 1996
    EventProceedings of the 1996 8th Mediterranean Electrotechnical Conference, MELECON'06. Part 3 (of 3) - Bari, Italy
    Duration: 1996 May 131996 May 16


    OtherProceedings of the 1996 8th Mediterranean Electrotechnical Conference, MELECON'06. Part 3 (of 3)
    CityBari, Italy

    ASJC Scopus subject areas

    • Electrical and Electronic Engineering


    Dive into the research topics of 'Intelligent forecasting of distribution system loads'. Together they form a unique fingerprint.

    Cite this