An Accurate Hybrid Approach for Electric Short-Term Load Forecasting

Abstract

For efficient working of the power system, an accurate approach for short-term load forecasting (STLF) is suggested. To improve the accuracy of forecasting, various weather conditions, such as temperature, humidity, dew point, wind chill, and wind speed, are considered and their impact on the accuracy of load forecasting is studied in detail in terms of Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Maximum Error (ME) errors. The proposed hybrid approach consists of Support Vector Regression (SVR) and fuzzy because SVR can forecast the ability of small dataset and fuzzy system to handle non-linear weather conditions and uncertainty of load in forecasting. For load forecasting, time of the day, historical load i.e. previous one-month hourly load, weather conditions, calendar days for the last 10 days, sunny time, temperature at the same time in previous day, and average temperature of last three hours are taken into account. The proposed approach provides accurate load forecasting for a day regardless of its being a working day or holiday, while fewer days are used for load prediction viz. previous one month, while no special care is taken for weekend. The suggested approach is tested on standard electricity datasets: EUNITE network 1997 and New England of America of 2012 and 2019. Simulation results show better effectiveness and the superiority of the proposed approach when compared with other existing methods for daily load forecasting viz. ANN, Bayesian, and Least Square Support Vector Machine, etc.