Neural data mining and modelling for electric load prediction

The authors examine the total daily load data for a large region of the UK over an eight year period. The objective is to examine the data and determine what factors influence the load level. The approach is to assume little knowledge of the system, starting with a minimal number of inputs and a network with few hidden neurons. This way the network will formulate a relationship between the given inputs and the load. By examining the peculiarities of those days which do not fit into the model it is possible to discover why they do not and to create extra inputs that convey the information required.