Connectionism-Based CBR Method for Distribution Short-Term Nodal Load Forecasting

Short-term nodal load forecasting is very important for operation, planning and management of distribution networks, which is characterized by lack and imprecision of historical data, inconspicuous trend for load variation, and changeful pattern of nodal load. It is very hard for conventional methods to solve such problem. A connectionism-based case-based-reasoning method (CBCBR) is proposed based on parallel distributed processing (PDP) model. The principle of CBCBR is analyzed, the elementary architecture of CBCBR is defined, and a hybrid supervised/unsupervised learning algorithm, which equips CBCBR with a good generalization capability, is also proposed. CBCBR can build nodes and connections dynamically by the rapid and incremental learning procedure and can withstand the effect of bad data effectively through network self-organizing. The proposed method is tested using load data of a practical system and the test results is compared with that of BP network, RBF network and AR model.