Prediction of power fluctuation classes for photovoltaic installations and potential benefits of dynamic reserve allocation

Abstract

During partly cloudy conditions, the power delivered by a photovoltaic array can easily fluctuate by three quarters of its rated power in 10 s. Fluctuations from photovoltaics of this size and on this time scale may necessitate adding an additional component to power system secondary and primary reserves to regulate frequency. This study quantifies the benefit of dynamically sizing a reserve component to cover photovoltaic fluctuations so that the additional reserves are different for each hour. The concept of categorising an hour as belonging to one of three possible fluctuation classes is presented. Based on historical array data and weather forecast information, several methods of forecasting these classes are evaluated, including persistence, Markov chains and a neural network. A practical tool based on class forecasting is proposed to aid in estimating a photovoltaic reserve requirement ahead of time for horizons ranging from 1 to 24 h. Results indicate that of a 10% possible reduction in total reserves held, most of this benefit (8%) can be obtained for hour-ahead scheduling with persistence forecasting and that a similar benefit may be possible for four-hour ahead scheduling if neural networks based on weather forecast information are introduced.