Application of least absolute value parameter estimation based on linear programming to short-term load forecasting

Short term load forecasting employs load models that express the effects of influential variables on system load. The model coefficients are found by fitting the load model to a data base of previous loads and observations of the variables, and then solving the resulting overdetermined system of equations. The coefficients thus obtained are critical to the forecasting process, as they directly affect its final predictive accuracy. This study compares two linear static parameter estimation techniques as they apply to the twenty-four hour off-line forecasting problem. Here a least squares and a least absolute value based linear programming algorithm will be used to simulate the forecast response of three twenty-four hour off-line load models. The three load models are (I) a multiple linear regression model, (2) a harmonic decomposition model and (3) a hybrid multiple linear regression/harmonic decomposition model. These models are simplistic in nature and their primary purpose is to provide a basis for comparing the two parameter estimation techniques. The results obtained for each estimation algorithm via each load model, using the same data bases and forecasting periods, are presented and form the basis for comparisons presented in the paper. (C) 1997 Elsevier Science Ltd