Short-Term Electrical Load Forecasting Using Support Vector Machines

For the purpose of short-term forecasting the paper uses one of the latest artificial intelligence methods – Support Vector Machines (SVM). Verification of the use of the model developed was carried out on the example of a section of the actual medium-voltage distribution network of the Novi Sad Power Company. The point of consideration was the impact that the weather factors and the number of previous days with their registered hourly consumption used as training data source have on the quality of short-term forecast results. Verification of the results was conducted by comparing the measured and forecast values which were based on three models: classical auto-regression, neural network and SVM. It was noted that SVM gave comparatively better results in short-term forecasting than neural network and better results than auto-regression models. Key-Words: Support Vector Machines, time series forecasting, electric energy consumption