An algorithm based on dynamic fuzzy granular neural network is proposed for the characteristics of power load,
which is time-variant, variable structure and non-linear. The quotient space theory of granular computing and fuzzy neural
network technology are used in power load modeling. Elliptic basis function and fuzzy ??
−completeness are also used as
a distribution mechanism of on-line parameter to avoid the randomicity of the initialization options. Online self-adapting
adjustment is executed on the width of input variable according to fuzzy rules and the importance of input variables, and
then the load parameters and the structure can be identified synchronously. Experimental results show the feasibility and
effectiveness of this method.