The electrical load, sampled every hour, at Salagatan 18 in Uppsala was used to form models and for forecasting the load. It was investigated whether Multi Seasonal ARIMA models and Support Vector Regression were suitable. The models were compared to a naive persistence benchmark in periods of high and low volatility. Both short range 24h and long range 168h forecasts were made. It was concluded that both model proposals could be used to forecast the electrical load series. ARIMA and Support Vector Regression model proposals outperformed the benchmark for long and short range forecasts in both volatile and non volatile settings. The mean absolute percentage errors of the best ARIMA model for a one week forecast were 15.1% and 21.6% for non volatile and volatile settings, respectively.
Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:uu-385594 |
Date | January 2019 |
Creators | Weinberg, David |
Source Sets | DiVA Archive at Upsalla University |
Language | English |
Detected Language | English |
Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
Format | application/pdf |
Rights | info:eu-repo/semantics/openAccess |
Relation | TVE-E ; 19006 |
Page generated in 0.0014 seconds