Artificial Intelligence for Energy Demand Forecasting

  • Alya Alhendi

Student thesis: Master's Thesis

Abstract

Accurate load forecasting is a significant step in the power system planning and operations. The application of intelligent approaches such as neural networks has been proved to deliver results with considerable accuracy. Two forecasting models for short-term energy demand forecasting are developed in this thesis: Artificial Neural Network (ANN) and Markov Chain based Artificial Neural Network (ANN-MC). The ANN model serves for comparison with the developed ANN-MC model. The models developed in this thesis consider statistical factors such as average daily load, average weekly load and date, as well as environmental factors such as dry bulb temperature and dew point, user behaviour aspects such as weekday and weekend. The models were trained using four different training algorithm to identify the most suitable algorithm for this application. The training algorithms are Levenberg-Marquardt, Bayesian Regularization, Scaled Conjugate Gradient and Grey Wolf Optimizer. The performance of proposed models for energy demand forecasting is demonstrated using different test cases derived from the given dataset. Different performance indices such as mean absolute percentage error, maximum percentage error, skewness, kurtosis and risk index are used for assessment. The performance of the developed model is studied on a yearly basis as well as seasonal basis. A risk assessment method based on the errors of the forecast is proposed and compared to another risk assessment method that is based on the standard deviation of the load.
Date of AwardMay 2019
Original languageAmerican English

Keywords

  • Artificial intelligence
  • artificial neural network
  • Markov chain
  • grey wolf optimization
  • load forecasting.

Cite this

'