Research Of Model Integration In Load Forecasting

The paper presents a new type load forecasting system based on data-warehouse, model base, knowledge base, OLAP and data mining. Model integration in model management has been studied in the paper that brings some theories about model integration forward. Those are the theory of model integration based on model manipulation, the theory of Error in model integration and the theory of optimization of load forecasting models linear combination. Correspondingly, the theories solve model integration, model integration selection and optimization of load forecasting model linear combination in the system respectively.Three requirments for model management in load forecasting are put forward firstly. Model management should do model integration in a flexible way. If there are many model integrations available, model management is supposed to select the best one among model integrations and do forecasting model linear combination, which is more exact.Chapter 2 to 4 present some solutions to three requirments. Chapter 2 presents the theory of model integration based on model manipulation. A formulized representation of model that includes input set, output set, and some concepts of model combination-relations and model integration results are proposed. In the following, whether or not model integration results exist is analyzed in detail, and several sufficiencient conditions are proved. Subsequently, the application of the theory in load forecasting system is given.Chapter 3 presents the theory of Error in Model Integration. Error is considered as one kind of criterion for calculation precision that is regarded as a gist for model integration selection. The paper puts the concept of the error of model integrations forward that represents calculation precision of model integrations. The computation of the error of model integration is provided also. Calculation precision of model integrations can be increased through it. Then, a method for model integration selection is gained. Like chapter 2, the application of the theory is also introduced.Chapter 4 presents the theory of optimization of load forecasting models linearcombination. All forecasting models are viewed as functions and form a linear space with load real value function. Further more, all forecasting models themselves form a sub-linear space. Endowed with inner product, the linearspace is a complete inner product space that is a Hilbert space. Therefore, minimizing vector theorem provides an academic gist that there must only exist a linear combination that is the closest to load real value function.Finally, the conclusion of the paper is given and the future work is suggested.