A Taxonomy of electricity demand forecasting techniques and a selection strategy
DOI:
https://doi.org/10.17722/ijme.v8i2.874Keywords:
Electricity demand forecasting, criteria for selection, stochastic time-series, ARIMA, Exponential Smoothing, Kalman Filtering, Artificial Neural Networks, Support Vector Regression, Expert systems, Genetic Algorithm, EconometricsAbstract
In this research, a taxonomy of known electricity demand forecasting techniques is presented based on extensive empirical studies. In addition, a decision strategy for selecting an electricity demand forecasting method has been presented. The strategy has been formulated based on an eight-factor model created by World Bank and inputs gathered from electricity demand forecasting experts (through a questionnaire). The techniques have been assessed based on time horizon, accuracy, complexity, skill level, data volumes, geographical coverage, adaptability, and cost. The experts rated ARIMA (Autoregressive integrated moving average) with exponential smoothing and Kalman filtering as the most adopted method. The next most adopted method is Artificial Neural Networks with preprocessed Linear and Fuzzy inputs. However, now Support Vector Regression may replace this method, which is currently tested by many electrical engineers engaged in electricity demand forecasting. In addition to these highlighted methods, this research also presents the ratings of other techniques based on the eight-factor model of World Bank.