The Application Of Non-Classical Econometrics In Load Forecasting

Load forecasting is of great importance for economic and secure operation of power system.With the development of electricity market, more and more attention is paid to the precision of load forecasting. There have been a good many efficient forecasting models in the field of load forecasting.But,unfortunately,some puzzles also exist.The preconditions of some classical models are hardly met in applications.And sometimes the parameters of some models have few definite physical meanings.This paper tries to model load time series with non-classical econometrics models, which have more practical preconditions, and have more powerful realistic meanings. The main achievements involve the following aspects:1. The actuality and challenges of load forecasting are analyzed.2. The autoregressive conditional heteroscedasticity (ARCH) model and cointegration theory of the non-classical econometrics are reviewed.3. The two-variable system, including seasonally adjusted power consumption and GDP, is modeled with dynamic economics model. With the help of cointegration theory, the relationship of cointegration between this two series is studied.4. The model of two variable system based on VECM is built.The long term equilibrium in this model is analyzed. The mechanism of error correction in short term is also discussed.5. The daily power consumption series is modeled with load-decompoing.The ARCH effect is analyzed. Finally, a generalized ARCH model is built.6. With the help of extended ARCH class model, a positive research on the second-order moment of load time series proceeds from different viewpoints, such as the influence of second-order moment upon the first-order moment, reverse leverage effect and modeling the long term volatility and the short term volatility together.7. Volatility clusrering effects of load time series in different terms are studied, and the intensity of ARCH effect in different series is compared.