Shortening time-series power flow simulations for cost-benefit analysis of LV network operation with PV feed-in

Time-series power flow simulations are consecutive power flow calculations on each time step of a set of load and generation profiles that represent the time horizon under which a network needs to be analyzed. These simulations are one of the fundamental tools to carry out cost-benefit analyses of grid planing and operation strategies in the presence of distributed energy resources, unfortunately, their execution time is quite substantial. In the specific case of cost-benefit analyses the execution time of time-series power flow simulations can easily become excessive, as typical time horizons are in the order of a year and different scenarios need to be compared, which results in time-series simulations that require a rather large number of individual power flow calculations. It is often the case that only a set of aggregated simulation outputs is required for assessing grid operation costs, examples of which are total network losses, power exchange through MV/LV substation transformers, and total power provision from PV generators. Exploring alternatives to running time-series power flow simulations with complete input data that can produce approximations of the required results with a level of accuracy that is suitable for cost-benefit analyses but that require less time to compute can thus be beneficial. This thesis explores and compares different methods for shortening time-series power flow simulations based on reducing the amount of input data and thus the required number of individual power flow calculations, and focuses its attention on two of them: one consists in reducing the time resolution of the input profiles through downsampling while the other consists in finding similar time steps in the input profiles through vector quantization and simulating them only once. The results show that considerable execution time reductions and sufficiently accurate results can be obtained with both methods, but vector quantization requires much less data to produce the same level of accuracy as downsampling. Vector quantization delivers a far superior trade-off between data reduction, time savings, and accuracy when the simulations consider voltage control or when more than one simulation with the same input data is required, as in such cases the data reduction process can be carried out only once. One disadvantage of this method is that it does not reproduce peak values in the result profiles with accuracy, which is due to the way downsampling disregards ce