Forecasting the unit cost of electricity generated by fossil fuel using ARIMA technique: A case study of a diesel/heavy fuel oil power plants in Sri Lanka

Abstract

—The national electricity grid system must be kept in balance between sufficient supply to meet the demand and minimization of the cost by alleviating excessive generation. The forecasts made from the historical electricity generation cost data can support the national grid system in this regard. Present study suggests a statistical time series model for forecasting the Unit Cost (UC) of electricity generated by fossil fuel power plants using Auto Regressive Integrated Moving Average (ARIMA) technique. It was conducted as a case study in a Diesel/Heavy Fuel Oil (HFO) power plant in Sri Lanka which consists of two sub power stations. The model was developed and validated using 80% and 20% of monthly data from selected power plant from January 2013 to June 2018. ARIMA (1,1,0) and ARIMA (2,1,2) were selected as the best models with the lowest Akaike Information Criterion (AIC) for Station 1 and Station 2 respectively, from among many candidate models that were evaluated by the investigation of ACF and PACF of the series. The forecasting accuracy of ARIMA (1,1,0) and ARIMA (2,1,2) models was measured with Mean Absolute Error (MAE) values (2.431 and 0.717) and Root Mean Square Error (RMSE) values (3.403 and 0.927).When comparing the UC of both stations, the forecasting values shows that UC of Station 1 are quite greater than Station 2 values and it is also relevant to past years’ cost data.