抄録
Recently, intense interest has been shown towards recursive algorithms for parametric identification of systems derived from Model Reference Adaptive Systems (MRAS identifiers) which assure global asymptotical convergence of solutions. We apply an MRAS identifier to one-day-ahead load forecasting. Based on actual data from Kyushu Electric Power Company and weather stations in Kyushu throughout 1981, the accuracy of the proposed load forecasting is found to be very high, the standard deviation of the error of the forecast being under 4%. There exists two important keys to the forecasting success of the MRAS identifier; the suitable construction of a load model and the proper use of an MRAS identifier.
According to properties of load curves, we construct 24 load models using hourly measurements of weather conditions during a 24 hour period. We then find the best load models, whose input signals are measurements of the daily variation of the discomfort indices and whose output signals are daily load changes.
Since an inadequate MRAS identifier causes poor identification of the load models, the selection of a good MRAS identifier is an important task. We select an MRAS identifier whose steady state gain is regulated by the two indices λ and μ. By using proper values of λ and μ, the MRAS identifier accurately identify the load models. Based on actual data, the pertinent values of λ and μ are found to be within the interval from 0.78 to 0.98.
