VSE knjižnice (vzajemna bibliografsko-kataložna baza podatkov COBIB.SI)
  • Comparison of linear, neural and ELM based models for short term heat load forecasting [Elektronski vir]
    Potočnik, Primož, 1969- ; Govekar, Edvard
    Various forecasting models are considered and compared for shortterm heat load forecasting in a district heating system. Heat load data and weather related influential variables for five subsequent ... winter seasons of district heating in Ljubljana (Slovenia) are applied in this study, and additional informative features are extracted to improve the forecasting accuracy. Forecasting models include linear autoregressive and stepwise regression models, neural networks and extreme learning machines. The models are developed with the objective to forecast the future daily heat load with the forecasting horizon one day ahead. Evaluation of the forecasting models is based on generalization error, obtained on an independent testing data set. Comparison of forecasting models reveals good forecasting performance of a linear stepwise regression model (SR) that utilizes only the most relevant input variables. The operation of SR model was improved by using neural network (NN) models, and also NN models with direct linear link (NNLL). The best forecasting result was obtained by using extreme learning machine (ELM) model. The results demonstrate the applicability of NN, NNLL and ELM models to accurately forecast the heat load data, but also reveal practical considerations in designing NN-based and ELM models. Namely, random initializations of NN-based and ELM models require multiple iterations of a learning procedure in order to achieve good forecasting results. Furthermore, ELM models are sensitive to the range of input variables because hidden layer weights are not tuneable but randomly chosen. Only if properly designed and trained, NN-based and ELM models offer a good forecasting tool for the district heating market.
    Vir: ITISE 2015 [Elektronski vir] : proceedings (F. 469-480)
    Vrsta gradiva - prispevek na konferenci
    Leto - 2015
    Jezik - angleški
    COBISS.SI-ID - 14083867

    Povezava(-e):

    http://lab.fs.uni-lj.si/lasin/pub/ITISE2015_proceedings.pdf

Vnos na polico