Study and application of data mining and NARX neural networks in load forecasting

The relationship between med-long term load forecasting and socio-economic indicators is very difficultly described by an accurate mathematical model. So load forecasting needs to dig out few dominant factors from lots of socio-economic indicators. By introducing data mining technology into the association analysis of China´s electricity consumption growth, many socio-economic indicators since 2000 are selected to constitute the relevant factors database. To complement of a few missing data, a number of indicators closely related to the electricity consumption are dug out by cluster analysis, and the data of distortion indicators are corrected, thus, a more scientific load forecasting model is built. The relation between electricity consumption and selected indicators is validated and tested by dynamic neural network time sequence tool. The results show that the prediction model has good convergence.