In wind integration studies, sub-hourly, load synchronous wind data are often preferable. These datasets can be generated by a hybrid approach, combining hourly measurements or output from meteorological models with a stochastic simulation of the high-frequency fluctuations. This paper presents a method for simulating aggregated intra-hourly wind power fluctuations for a power system, taking into account the time-varying volatility seen in measurements. Some key elements in the modelling were transformations to stationarity, the use of frequency domain techniques including a search for appropriate phase angles and an adjustment of the resulting time series in order to get correct hourly means. Generation data from Denmark and Germany with 5 and 15 min temporal resolution were used for training models. It is shown that the distribution and non-stationarity of simulated deviations from hourly means closely follow those of measurements. Power spectral densities and step change distributions agree well. Of particular importance is that the results are good also when the training and objective power systems are not the same. The computational cost is low in comparison with other approaches for generating high-frequency data.
NOMENCLATUREVariables in continuous time are denoted with the time argument in parentheses, while discrete time is indicated with subscripts. Additional subscripts indicate whether the time series are simulations or measurements used for validation of the model performance, e.g., P t,sim and x t,val . Subscript obj is used for variables that are common for the objective power system (i.e. both simulation and validation). No additional subscript indicates training data or when the variable is discussed in more general terms. The first-order and second-order derivatives are denoted with prim and bis, mean values with a vertical bar and absolute values with horizontal bars.