Determination of similar days in load forecast based on Grey incidence theory

Next-day load curve prediction is an important module of electricity market operation system. Artificial neural network (for short, ANN) is one of successful algorithm used in shot-term load forecast. But one of key issues is how to select the trained samples and input variables, which will influence the convergence, calculation speed and calculation precision of the model . Based on analysis of daily load data and daily meteorological data in Changchun, we proposed the set of multi-interval meteorological data is considered as the criterion of determination of similar days. With Grey incidence theory, the incidence degree of temperature variation curve in history days and that of forecast days is obtained. Then the days, which have similar meteorological condition to the next day to be forecasted, are selected as the samples of ANN. In this case, the model of daily electricity forecast, which has good adaptive characteristics, is constructed. By wave coefficient of daily load curve, the estimated load at 96 intervals in one day is gotten. Verified by continuous prediction for one month in practice, the result is satisfactory.