Research Of Cross Weight Changeable Combined Power Loads Forecast Model

The load forecasting is an important routine for power dispatch department. Especially with the foundation and the development of power market it will bring into play a more and more important role. It's precision directly influences power system's security, profit and quality. Therefore, how to improve the forecasting precision is the emphasis on the study of load forecasting.Combined Power Loads Forecast is an important research direction because it can reduce the risk of load forecasting. The paper first Research the existing combination forecasting model, including the optimal combination forecasting model and non-optimal combination forecasting model. Existing combination forecasting model can deal with the problem of weight distribution, but there are some unreasonable. For example, they can't deal with the fact that the impact of different history load to the forecasting load is not the same. It also can't save the problem that different forecasting methods have different contribution to the final forecasting load at different time. After studying the existing forecasting model, the paper give out an abstract optimal combination forecasting model and prove that existing models is concrete examples of the abstract model.In order to improve the shortcomings that existing models can not deal with the changeable impact of history load to forecasting load, the paper instance a new model base on the abstract model. In the new model it make different history load have different weight. Then it gives out two methods to calculate the weight. In the first method the weight depend on the date. In the second method the weight depend on the forecasting error. It sorts the errors and then determines the weight according the sequence results.In order to improve the shortcomings that existing models can not deal with the changeable impactions of different forecasting methods at different time, the paper give out a matrix. So the existing model weight matrix is a special matrix that very line in the matrix is the same. In the new model the matrix is calculated as follow. It first makes very line in the matrix the same, then it adjust very line according forecasting error.The paper presents the final model that deal with the two unreasonable. Results show that the model improved the prediction accuracy.