Monthly electric load forecasting using GMDH

A forecasting method with their objective representations is proposed. A trend of monthly load data, as well as the remained stationary component, is modeled by use of GMDH. Then forecasts are calculated using the models. Owing to the feature of GMDH as a heuristic self-organizing method, models just fit for the data in hand can be obtained without any modifications of the modeling algorithm, even if the underlying characteristics of the data are changed. Therefore the method can be applied in various situations with observed data containing different nonstationary characteristics. Results of the application to several sets of load data in Japan with and without abrupt changes demonstrate the advantages of the proposed method.