Sizing a flexible spinning reserve level with artificial neural networks

The paper presents a neural network based method to formulate an economic flexible spinning reserve strategy, so that more reserve is withheld when contingencies are likely to occur, and less reserve is kept otherwise. The method firstly employs a Kohonen neuron network (KNN) to analyse historical data, such as the hourly spinning reserve, hourly utilised spinning reserve, and the hourly demand data. This data is clustered into different groups, where they could be reserve over-held, under-held or satisfactory to different degrees. This is followed by suggestions of appropriate reserve level for each of the groups. Based on the analysis results and the suggested reserve level, the paper then trains a fuzzy neuron network to determine future spinning reserve so as to both minimise operating costs and enhance system reliability